TWI280488B - Online intelligent information comparison agent of multilingual electronic data sources over inter-connected computer networks - Google Patents

Abstract

A method and system for real-time online search processing over inter-connected computer networks, in which an offline database information is maintained for a plurality of vendor sites from the inter-connected computer networks. The information includes URLs, search form URLs, description of domains, and vendor descriptions, and while the vendor descriptions comprise generalized rules about how product information is organized on each of the vendor sites. Parameters are processed for a price comparison request for a desired product using the information maintained in the offline database whereas price comparison request is received from an online user or buyer and/or from the system of the present invention. Real-time price and product information is then extracted from identified ones of the plurality of vendor sites, wherein the extracted price and product information are in a native language of the site; and the extracted price and product information are displayed to the user.

Description

1280488

V. INSTRUCTIONS INSTRUCTIONS (1) The provisional application No. 60/236,574 filed on September 29, 2000 and No. 6/299,360, which was broadcast on June 19, 2001, is at 35u.sc§119 ( e) Declare priority. The list of attachments is a list of computer programs on a total number of discs (including a piece of copied disc). A single file included in the proposed optical disc, named ''implement.txt', was created on September 18, 2001, and its size is 14K bytes. The material contained in the optical disc is thus It is hereby incorporated by reference. The invented is generally about automated work on the World Wide Web (Web), more specifically about online buyers or users through a single interface and global information online multi-language sellers. Automated work, such as comparing shopping or talking, to improve communication efficiency and provide personalized purchasing experience. Since the establishment of the World Wide Web in the mid-1990s, the size of the Internet has expanded by a thousand times. People are no longer face to face. The conversation is interconnected, but through a virtual communication channel. The new revolution in this technology has fundamentally changed the way people live. Parallel to the global information network is the "information technology generation", which represents an amazing variety of online information resources, The scope includes product information to academic papers. These elements have contributed to the exponential growth of e-commerce, which capitalizes on the convenience and low cost of the Internet. There are millions of online sellers on the Global Information Network. Although the current comparison shopping or price comparison search engine can obtain the search results related to any desired product and its desired price requirements by different global information network competitors according to the requirements of the buyer or the user, the buyer or the use Meeting 1280488 V. Invention Description (2) There is no information on the sea. Sometimes, because the search engine misses the website of the existing online multilingual seller in the other Internet-connected countries (currently 245) that are being sold, the buyer or user will receive the "webpage". "invalid" search results. Also, although the information about the product and the seller is easily accessible on the Web, the buyer or user will still be in the loop of all stages of the purchase process. The potential of the Internet to transform e-commerce's current modality into a real global market is still largely unrealized today, and e-shopping is still not automated. Shopping on the Internet is far from simple, effective, or enjoyable. The Bayer Engine and the centralized directory service are inefficient in locating the products that online buyers want and the merchants who are willing to sell those products or services. Also, typical online shopping programs are mostly manual and require the buyer to type in all the items and keywords they are searching for. Therefore, potential buyers will face the loss of confidence and will be responsible for collecting and interpreting information about merchants and products. The situation is that the user or the buyer will not have enough time and expertise due to excessive information load. In terms of complexity, the two imperfect strategies currently implemented to automate the online catalog price comparison request processing are as follows: (1) Non-immediate practices (2) Instant hard-coded packages are not immediate practices. The easiest way to make a price comparison agent. In fact, it consists of manually collecting all the necessary information by the Web, and then writing separate HTML files for each item of the search results to visually display the search results. 1280488

V. Description of invention (3) The benefits of the above are obvious. One implementation is easy and the search time is short. Despite these benefits, there are three major undesired deficits. First, price comparisons are done manually, and maintaining large parcel storage becomes very costly, especially in the case of continued growth of the Internet. Second, it is necessary to invest in time to update prices and other information. Finally, the database needed to store and coordinate the above information can be very large. The instant hardcode parcel approach is an alternative to non-immediate practices. Instead of getting the project directly from a non-instantaneous approach, the immediate practice attempts to generalize the HTML page into a specific format. To perform this extraction, a custom package called pcwrapHLRT (program prefix) is fired. Figure 1 provides an example of a fixed portion of a program with a '' While' loop. In this example, the rule after creating a package is to use a pair of tags to define the target material on the HTML page. Because the site exhibits uniformity Format agreement, this pcwrapHLRT program will succeed. Product items are presented in bold type and price in italics. pcwrapHLRT by "<b>,,,*'</B>'," <I>", "</1>"} scans the HTML file and determines the paragraph of the text to be extracted. These strings are respectively defined by pcwrapHLR as I, η, Ip and rp. This notation Ik(ke{i,p}) indicates the left side of the attribute to be extracted, and rk indicates the right mark. Other attributes that can be extracted by the package are product names, graphics, items and conditions. When an HTML web page is given, pcwrapHLRT starts scanning the entire web page sequentially by the header number. The peripheral loop checks whether additional model numbers and/or price pairs are to be extracted by scanning the "<B>" flag on the non-scanned portion of the web page. As soon as a model number is found, the inner loop is 1280488

V. Description of the invention (Ο was motivated to extract the appropriate page substrings. Few websites publish their format agreements. Therefore, those who use the pcwrapHLR Dingzhi information collection system will manually build this package for each resource. Unfortunately, this hard-coded process is boring and error-prone because ordinary HTML pages contain thousands of lines of programming. Moreover, most websites periodically change their format conventions, which usually breaks the package.另一. Another disadvantage of pcwrapHLRT is that the agent must contact the seller's website when receiving the request from the user, so the search time is quite speedy. Since the package is partially automatic, additional administrative work must be done. It is implemented to manually analyze the format of the HTML to determine the package. In view of the aforementioned problems encountered in this common encounter, an alternative approach is to manually and partially automatically manipulate the online intelligence according to the new Internet strategy. Type price comparison agent, which can slow down the online catalog purchase or purchase price comparison processing (auction), and can be related to the Internet The network interactive agent feature graphical user interface (IACGUI) also provides a better navigation environment. This is in the so-called fourth-generation global overall market architecture - agent arbitration B to c, c to C, B to B , e-procurement and auction, G-to-B/C (government-to-business/consumer) bidding e-commerce and m-commerce (mobile commerce) have become particularly important when implemented. Preferably, the online smart price comparison agent is an automated purchase or shopping assistant that searches for global multi-language stores for each product and explores the transaction. It also delivers a bonus (customer rating) The company's global information network service to online buyers/users. This kind of agent is attractive, and it can be mitigated by making it 1280488. 5. Inventor's Note (5) The user is manually implemented in the consumer purchase behavior model. Difficulties in each job. Conventionally, the buyer/user communicates with the web server of the online service through the front-end interface, which presents the search request by the buyer/user, and the web server of the online service queries its database to The mediation and presentation of the result to the user's web browser. In the present invention, the user agent (online intelligent price comparison agent acting as a human buyer/user) is online in the catalog price comparison process. Many multilingual web servers with any one of the countries from the country to be connected to the Internet and the Internet in the best interests of the buyer/user in the best interests of the buyer/user. The user agent then ranks the online seller websites it finds and presents the search results to the human users on the line via a web browser. The benefits of applying the system of the present invention to multiple e-commerce segments are very significant. Communication efficiency and effectiveness are significantly improved, and time and cost savings for online vendors and online buyers are maximized. Most importantly, users/buyers will have access to an unprecedented number of sources of information on a global scale with tens of thousands of products and unmeasured business opportunities. The system and method of the present invention will also help to smash time and language barriers, demographic boundaries, and truly contribute to the globalization of e-commerce. In addition, the personalization, continuous operation, and automatic nature of the user agent make it quite suitable for arbitrating buyer/consumer behavior. Its invention will help to optimize the entire purchasing experience and the current revolution in e-commerce. It is therefore an object of the present invention to provide an improved online seller of 1280488. 5, Invention Description (6) Price comparison of products or services. Still another object of the present invention is to construct a vendor description for an online store. Still another object of the present invention is to collect data, including product samples and URLs that are used for training. Still another object of the present invention is to obtain training materials prior to training on a vendor website or online store. Still another object of the present invention is to collect training web pages from online vendors using the information given in the training materials. Still another object of the present invention is to generate a vendor repost from the training material and the collected training web page. Still another object of the present invention is to store the resulting vendor description in an offline database. Still another object of the present invention is to provide a system administrator with an interface to add, modify, and delete vendors that are supported by the system. Still another object of the present invention is to provide an interface for an administrator to view vendor information. Still another object of the present invention is to provide a price comparison method by which a hacker can initiate price comparisons. Still another object of the present invention is to analyze the HTML web pages into useful materials. Still another object of the present invention is to filter and sort desired products/services. Still another object of the present invention is to provide a single interface for comparing the price of different online multi-language vendors and different domains on the Internet or the World Wide Web. 1280488

V. INSTRUCTIONS (7) In the system of the present invention, a first user agent is implemented and is implemented in the form of a Semantics Rec〇gniti〇n Learner a 2eni SRLA Work. It uses the inductive learning method to perform the instant automatic parcel induction to the URL of the seller's website and its domain description according to the compiled or prepared training case previously provided by the system administrator. (In a real shot, SRLA connects to the Microsoft backend. Obey SQl's server or Microsoft Access Library to generate a vendor and product description for each online store). The parcel induction is accomplished by building a package of instant examples that are extracted by the vendor and product description stored in the offline repository. Then, srla automatically uses the examples to quickly and quickly drive to the remote host of the seller's website through the Internet to access a webpage displaying a specific example according to the provided URL, and then intelligently fill in the relevant information by the domain or product information. Search the form and then "press the enter key" to make a search request for the site. The results of the web page returned in response to the search criteria are successful or invalid pages with the correct information. These web page results with unique vendors and product descriptions belonging to a specific vendor (whether or not there is a login in the system) are then stored in an offline database (such as a server compliant with SQL or a Microsoft Access Library). The queue is maintained by the system administrator. The vendor URL, vendor description and other information are preferably automatically updated on a daily basis according to the schedule. In the system of the present invention, a second user agent is referred to as a Semantics Recognition Buyer Agent (SRB A). SRB A uses the semantics previously learned by the learner agent. 10 1280488 V. Description of invention (8)

Vendor descriptions are used to search for compliance, while simultaneously accessing various online multilingual seller websites on the World Wide Web. SRBA intelligently fills in the seller's search form with the product information provided by the online buyer or user and virtually "presses the enter key". The seller then sends back the search results page to the SRBA via the World Wide Web, in that the resulting web page arrives at approximately the same time as the other sellers are returned. (This semantically distinguishes the buyer agent from storing these returned web pages in separate memory or cache memory for hitting to be used later by the SRBA). The SRBA analyzes the returned web page according to the corresponding vendor description, extracts related information and materials, sorts the price and model, and displays it on the client machine to the online buyer via a web browser with a formatted summary. /user. In accordance with the present invention, a method is to resolve the learner agent in a computer-implemented manner to provide inductive learning. The method includes mining training materials based on online vendors to generate corresponding vendor descriptions from interconnected computer networks. The method includes collecting training web pages using training web pages that are provided with training materials stored in the vendor queue. The method of using the training material and the training page taken by the recruiter includes an inductive learning method to generate a vendor-specific vendor description from the training material and the information extracted from the captured training web page. A method is provided for The retrieved and/or extracted vendor description is stored in an offline repository, which will be used later by a semantically distinguishable buyer agent (SRBA). One method is provided in accordance with the present invention for price comparisons of products or services by an online seller. The method includes an online user initiating a request for a particular product, or a service, and then semantically distinguishing a buyer agent using a predefined vendor description to construct a search request parameter. The method includes mailing the request to a different online vendor, and preferably extracting the data from the results web page returned from the online vendor using a parser containing the vendor descriptions. The method includes semantically distinguishing the buyer agent to construct/combine in HTML format. The sorted and filtered material is used to present the material to an online buyer, user. One method is to distinguish between a buyer agent and a proxy for being used to parse a web page sent back by an online seller to retrieve useful information. The method includes extracting the seller description from an offline data, parsing the webpage returned by the online seller for any country connected to the Internet on the global information network (currently 246), and using the returned The information described by the seller describes useful information. In one embodiment of the invention, the above functions are only available on the online web page of the buyer after logging in as a temporary trial or a lifetime member. In accordance with the present invention, a method is provided for instant online search processing of selected information patterns on interconnected computer networks. The method comprises a number of steps: combining a plurality of website descriptions on a plurality of connected computer networks, each of the plurality of websites including (a) a URL for the website; a search format URL used by the website; b) the generalization rules of how the selected information type on the website is organized; (c) the sample data retrieved from the website corresponding to the selected information type; and the description of the domain found on the website; A specific type of information received by an online user. 12 1280488 5. Invention Description (i〇) request; such websites describe the website with the specific information type; use the website description for each identified website Build a search request for the particular type of information; submit a search request constructed to the identified website; receive the search result from the identified website; and (d) at the time the received search result is to be correctly matched The local language of the website extracts information corresponding to the particular information type and displays the extracted information to the user. More generally, the present invention relates to a method for instant online search processing on interconnected computer networks. The method includes the steps of: (a) maintaining an offline database information for a plurality of vendor websites connected by the interconnected computer networks; the information including a URL, a search form URL, a description of the domain, and a vendor description. The vendor description includes generalization rules regarding how product information is organized on each of the vendor websites; (b) using the information maintained in the offline database for the desired product for price comparison requests to process parameters, And the price comparison request is received by an online user and/or semantically distinguishing the buyer agent; (c) extracting real-time price and product information from one of the identified ones of the seller websites, wherein The price and product information are taken in the local language of the website; and (d) the price of the product and the product information are displayed to the user. \ Figure 1 is an example of the permanent part of the pcwrapHLRT program that is used to extract information from the \ main website in the case of the conventional instant hard-coded package. Figure 2 is a generalized embodiment of the present invention, a user agent of the present invention, a user/buyer and an online seller via the World Wide Web/Internet. Figure 3 is a simplified flow diagram 100 of how the semantic learner agent (SRLA) acts as a training vendor 13 1280488. Figure 4 provides a description of the types of information that can be included in the vendor description in accordance with the present invention. Figure 5 is a diagram showing an example of information that can be resident in the vendor description in accordance with the present invention. Figure 6 is a simplified flow diagram of the semantically-resolved learner agent (SRLA) that implements inductive learning and produces a generalized web page effective vendor description. Figure 7 provides an example of a rectification portion that appears on a web page of a website, such as a person browsing the web page on the Internet, and an HTML code that is used to generate or define the rectification. Figure 8 provides an example of a label that is used in accordance with the present invention to determine the location of item descriptions and price information in a training web page. Figure 9 is a generalized description of the ones represented by the markers used in the training program in one example of the present invention. Examples 1A and 10B provide examples of possible candidate markers in the training process examples described in Sections 5 through 9. The figure is a simplified illustration of a single frame of a web page having navigational rules in accordance with the present invention utilizing searchable metrics and product domain fields for easy access to particular query data. Figure 12 provides a simplified depiction of a single web page of a web page showing uniformity in a simple and consistent format for all items. In the ··· Week page is a box, and the box contains the search results of the information queried, and these results are uniformly formatted.

14 1280488 V. INSTRUCTIONS (12) Figure 13 is a simplified description of the same single frame as shown in Figure 12, which shows the arrangement of products using search results using vertical isolation rules, centered between the head and tail. . Figure 14 is a generalized illustration of the semantics of the present invention. Figure 15A shows the search results on the seller's website "www.8〇〇.com, using "electronics" as a keyword, each of which has a description of its features and functions (left and middle of the arrangement box), and The related "listed price" and "your price" information appear to the right of the arrangement box, and the semantics of the present invention distinguishes the intelligent price identifier of the learner agent from the learning process described by the seller. 15B is a generalized illustration of the semantic distinguishing buyer agent of the present invention. It accesses the seller website "www.800.com, after the learning process described by the seller as shown in Fig. 14. 苐15C A flow chart for the semantics of Figure 2 to distinguish the buyer agent (Srba) 20 from the vendor description in response to a request from an online buyer/user for an online multi-lingual vendor that is all available for comparison Figure 30 is an example of an interactive agent character learner interface screen used to obtain training information in the present invention. The seventh information is that the training information has been sold to the seller "lcachec〇m," Into the example. Figure 18 shows the learner interface screen used to display the vendor description information that has been learned. 15 1280488 Factory V. Invention Description (η) Figure 19 is a single screen of the learner interface with the label “Seller Information”. The seller information can be searched by the key person or by. Figure 20 provides a single picture of a learner interface previously entered for a particular vendor to display such training examples. Figure 21 is a single screen of the learner interface displayed in response to the opening of the file called "Seller Description". Figure 22 shows the selection learning option in accordance with the present invention, that is, the "learn one" option is not selected, and the name of the vendor (which has been filled in) is "1 cache.com". Figure 23 shows the learning result of the seller "icache c〇m," Figure 24 shows the package induction problem, which is defined by the definition of a simple model for information extraction. Figure 25 provides the virtual code of the program "exccHLRT". Figure 26 is a block diagram of a simple method for learning the head and tail marks of the virtual code. Figures 27A and 27B provide a detailed table and associated subroutine for the program leaniHLRT. Figure 28 shows One embodiment of the present invention shows how a user/buyer communicates with a server to execute a processing DLL building on a server machine via an ASP (Active> (Web) website (NextGen.asp) (NextGen .dll). Figure 29 shows a semantic method for distinguishing the buyer agent from facilitating communication between the user and the database server. Figure 30 provides an illustration of how to set up SQL Server 16 1280488 in accordance with an embodiment of the present invention. Detailed flowchart of the ι server. Figure 31 shows semantically distinguishing how the buyer agent virtually mails a search back and forth on an online seller website. Figure 32 is a GUI or interactive agent character used in the present invention. A simplified illustration of such offenders / buyer interface (IACS / BI) in the "Main Menu" screen of its upper right corner to note that in this "Main Menu" screen of the products have to select "Frequency

"道" (category) is provided for this user. A "fast screen" is also provided on the left side of the screen. There is a box directly below it, which has its own animated appearance to indicate how online human users use the quick search option. The left panel also provides a set of boxes for members to log in as a temporary trial or lifetime membership. (Note that most of the entry functions of the present invention are disabled until the user recognizes that the certificate is valid). In the lower left corner, a set of online sellers connected to the portal of the present invention is provided, and on the right side, a large message box is marked as "feedback". Online users are provided to serve the e-mail through e_mail. The button has an annotation message, which is preferably executed using Microsoft's Outlook Express email application. Figure 33 is a simplified diagram of the GUI for the GUI or shopper/buyer interface of the present invention, in which the company responds to the "government-to-business" text that has been selected by the online buyer/user in the previous screen. The display is shown below. However, because these companies, or so-called government-to-business e-business services or providers, strictly restrict access to their web servers by adopting a certification security interface in a completely closed computer network environment. The right, this real picture does not work. The 34th circle is used for the Gm or shopper/buyer interface 17 1280488 of the present invention. 5. Description of the invention (15. A simplified illustration of the screen, wherein the user has clicked on the picture on the 33rd screen. The Step Search option is provided by the details of the company selected by the user in the provided selection. Note that in this screen, the headings in the panel box directly below the labels of the five domains are the r-level agents. Except at the bottom of the screen, the user is provided with a dialog box that can be populated to perform a search using the semantics provided by the present invention to distinguish the buyer agent. However, again, due to these companies, or so-called Government-to-business e-business services or platform providers are now strictly restricting the right of members to access their web servers by adopting a certification security interface in a completely closed computer network environment. This real picture does not work. Figure 35 is a simplified illustration of a screen for the Gm or Shopper/Buyer interface of the present invention, in which the company responds to a "business-to-business" that has been selected by an online buyer/user in the previous screen. The text is shown below. However, these companies, or so-called business-to-business e-business services or platform providers, are now strictly limited by adopting a certification security interface in a completely closed computer network environment. The member's right to access his web server, this real picture does not work. Figure 36 is a simplified illustration of the display screen for the GUI4 shopper/buyer interface of the present invention, where the user is already in the pictogram The details of the company selected by the user in the provided selection are provided after clicking the "Advanced Search" option on the screen. Figure 37 is a display screen for the (10) or constructor/buyer interface of the present invention. A simplified illustration of the selected item and its description is displayed in response to the user selecting "Domain A" on the screen. 18 1280488 V. Description of Invention (16) Figure 38 is a GUI for use in the present invention or A simplified illustration of the display screen of the shopper/buyer interface, in which the listed sellers are selling items in the domain a in response to the user having selected the "Advanced Search" option on the screen of Figure 37. .

Figure 39 is a simplified pictorial illustration of a display screen for the GUI or shopper/buyer interface of the present invention in which search results details that are characterized using the semantics of the present invention are provided. The shopper/buyer responds to the search request submitted by the user through the search parameters displayed at the bottom of the screen in Fig. 38. Referring to Figure 2, a generalized diagram is provided to illustrate the interaction between a user/buyer 12 and an online vendor 14 via the World Wide Web/Internet 16 in accordance with a preferred embodiment 10 of the present invention. In a preferred embodiment 10 of the present invention, a learner agent 18 (also referred to as a semantically resolved learner agent) and a shopper agent 20 (also referred to as a semantically distinguishing buyer agent) are provided. . A server 22 is utilized to provide access to an offline repository 24 that stores global multilingual vendor information. A system administrator 26 prepares/compiles the training material for the selected vendor website and stores it in the vendor list "27" of the offline database 24 via the server 22. The system administrator 26 can then use the training information to identify the learner agent 18 to "inductively learn" the training web page that has been retrieved from the main website via the global information network 16. This "inductive learning" forms a vendor description in the form of a vendor description, which is stored in the offline database 24. The user/buyer 12 can use the preferred embodiment of the present invention to retrieve the designated subject matter of the designated subject by semantically distinguishing the buyer agent (SRBA) 20 from 19 1280488 5, the invention description (l7). News. The SRBA requests processing by the user/buyer 12 by using the information contained in the previously described vendor description 28. The information in the Vendor Description 24 allows semantics to distinguish that the Buyer Agent essentially uses the World Wide Web 16 to instantly prepare and issue searches for many seller websites. The vendor description also allows semantically distinguishing the buyer agent 20 to process the received search results instantaneously and present the results of the filtered and unrelated information from all of the seller websites being searched to the user/buyer 12. Referring now to Figure 3, a flow chart 1 shows the operation of the semantically resolved learner agent (SRLA) 18 of an embodiment of the present invention. In a preferred embodiment of the invention, the semantically resolved learner agent 18 is implemented within a computer program executed on a server or personal computer. In step 11, the semantically determined learner agent is retrieved from the "seller's queue" 27 stored in the training database 35 to obtain training materials that are previously defined or prepared earlier. The training data 庠 24 is preferably offline. The training material includes a bundle of information belonging to online sellers whose information will be learned. These information can be displayed as shown in the right-hand column below, including URL, domain description, product samples and attributes, and other information depending on the domain. 卖i vendor name circle—, — 1 cache.com vendor URL http://www .cache.com Vendor search form URL http://st4.yahoo.com/cgi- bin/nsearch?catalog= 1 cached querv= Learning domain dvd 20 1280488 V. Invention description (1S) Training example---- 1 i) DVD virtual notebook theater with I-Glasses ii) JVC XV523GD Dolby Digital DVD player iii) PianeerDVL_919·Combined LC/DVD/CD player

Figure 4 shows an example of the type and description of the name of the post-training or post-learning data in accordance with the present invention. Figure 5 shows the actual "data element" of the learned example, which was generated when the vendor described the learning process in Figure 4 and was stored in the vendor description of the offline database and was used by the system administrator 26 maintain. The "post-training" information is preferably stored in a compliant SQL or Microsoft Acess database. This adds scalability to the choice of data containers from different vendors. Typically, the post-training data is independent of the online vendor's product domain, write characters, and presentation style. The exception is the URL path within the post-training data, which is required to uniquely identify different vendors. Returning to Fig. 3, a check is made in step 来看20 to see if the seller still has to be semantically distinguished by the learner agent 18. If there is still a seller waiting to be learned, the semantically resolved learner agent will proceed to step 13: No, the sub-process terminates. At step 130, the semantically determined learner agent 1s intelligently accesses the particular learning vendor corresponding to the predefined training material using the pre-defined training material. The semantically determined learner agent 18 searches for the particular product via the vendor website for the product identified in the training profile. Typically, the semantically resolved learner agent 18 is referred to as a "training webpage" by the system of the present invention or manually by the system administrator to retrieve training webpages, which will be used later to implement induction. Learn. In 21 1280488 V. DESCRIPTION OF THE INVENTION (19) In the preferred embodiment, the control information (the training material that would trigger the error page of the seller's website) is also included at this stage. Next, in step 140, the computer program performs inductive learning on the training data obtained by the semantic learner learner agent 18. The goal of inductive learning is to obtain a description of the attributes of the website and how it organizes product information and logically presents the product information to potential online customers. This learning product is referred to as the “Seller Description” and will be further described and explained at this stage with reference to Figure 6. Then in step 150, the semantically resolved learner agent 18 preferably selects the learned picture in the compliant SQL or Microsoft Acess database 24. (The vendor information or "seller description" 28 selected in the offline database 24 will be used later to distinguish the buyer agent 2 from the online). After the completion of step 15〇, the learner agent 18 is returned to step 12 to see if there is still a seller to be learned. If so, steps 13〇 to 150 are repeated. Otherwise, the learning process is terminated. Returning now to Figure 7, the vendor description learning process will be explained in further detail using a simple model of information extraction and a simplified training webpage example. The left side of page 7 shows the model arrangement and price information that will appear when potential customers browse the seller's website. For example, the top three behaviors on the right side of the html code provide the name of the arrangement, "Simple Product Catalog," and indicate the beginning of the information to be displayed. The fourth line provides the title text of the table - "md price". The sixth and seventh lines provide the name texts for these columns - "Model" and "Price". Models 8 and 11 provide model and price information. The rest of the information ordered is the end of the form, the arrangement of the table, and the end of the product catalog. '

1280488 V. INSTRUCTIONS (20) First, a package function produces a set of "markers" on the training pages given. The marker is used to locate the information of the training product in the training webpage. Figure 8 shows a simple product training webpage of Figure 7, which is a set of logos generated by the semantically resolved learner agent 18. The label j in Figure 8 indicates that the simple catalog page of Figure 7 contains four "tuples" where each tuple contains an "item" value and a "price value. A pair of integers represents each one. Considering the first pair, <174, 180 > These integers indicate that the attribute of the first tuple is a substring between positions 174 and 180, i.e., the string 'HM 381MD' is located between position 174 and position 180. In this case, by the user, the position is the number of characters from the start of the designation, such as the beginning of a web page or the end of a "head" of a web page. Spaces between text characters are counted as a character position. The letter ''H' in the check mark of the 7th figure "HM 381MD" appears in the first line by 174 characters from the character; and the ''D' in the string appears at the character position 180. Similarly, the last tuple or pair < 350, 301 > indicates that the last attribute price appears between character positions 356 and 361 and recognizes the string "399.95". It will be understood that although the character position is here The examples are used to identify "marks" and other criteria can be used within the spirit of the present invention. Other criteria can be used for the application of the methodology of the present invention. For example, check the seventh point again, and consider the semantics of the second language. The learner agent 18 autonomously assigns values to the model and "your price" for the four electronic products, and it is expressed as the following formula: Degree price! 28〇488

V' invention description (21) MD2070 599.95 MD203 249.95 MDR3 399.95 L= <b2,i,e2,i>,<b2,p,e2,p><b3,i,e3,i> ' <b 'P,e3,p>

^ <b4,i,e4,i>,<W,P,e4,p>J If "b" represents the start and "e" represents the end, the value identifying the position of the second tuple contains the string b -, i, the value at the beginning of the model "M", and the string e-, i is the end of the model. The analogy is that it will be understood by

Inspired by a standard relational data model containing the "item identifier" and "smart price identifier", a module's heuristic search facilitates the automation of the label, where the inverse tuple is a vector of two substrings &lt; b2,i,b2, p>. The string b—, i is the value of the item attribute, and the string b_, p is the value of the price attribute. So, the attribute represents the row and the tuple represents the column. "b2,, and the value "b2, i" between ",," means the position on the second column - the position value (mark) is calculated so that the entire Ppc (page of the product catalog - web page "p" is included The seller of the desired information describes the effective and learnable package induction that is implemented immediately, automatically, and hurriedly, regardless of the seller's website (in this case, ^ ♦ 8 0 0. com)# The formatted web page is a local character string for any language or is encoded in HTML, XML, cXML, Java, etc. or natural language. The indication of the content of the training web page is presented more generally in section 9. The information on the broken line is identified. In this case, the product and price are the information being searched for. In the second line, the "markers" are marked as "tuples". The corresponding ones are &lt;product left mark, product right mark> and <price left mark' price right mark &gt;. After the system administrator has executed the learning system once, then it compiles the possible candidate tag groups in parallel with the candidate tag groups in parallel with the possible candidate tag groups in the 10A and 10B diagrams. The seller browses the training page. And as shown in the above example, the other set of training web pages is continuously used in the instant, rush calculation including the temporary automatic real (signal) of the position value. Since the intersection of the two sets of candidates produces a valid candidate set, the semantically resolved learner agent 18 selects a valid candidate set as a vendor description. Referring now to Figure 2, a flow chart 200 illustrates an embodiment of the semantic discriminative learner agent 18 of the present invention. The methodology uses three environmental regulations 'which manages the configuration of the product descriptions provided in the shopping website, which allows the information extraction to be performed independently of the domain. These laws include navigational rules with searchable indicators, uniform laws, and vertical isolation laws. Regarding the navigation rules, online stores or seller websites are designed to serve consumers and business buyers. As a result, almost all online vendors provide searchable metrics for easy access to specific queried databases. Using the searchable form of the vendor website facilitates the semantics of the learner agent 18 to generalize the format of the multilingual home page and web page. Figure n shows the simplified map showing the top page with searchable metrics and product domain columns. Regarding the uniform law, although online stores or sellers differ widely from one another in their product description format, any known online seller typically arranges all project descriptions in a simple and consistent format. Figure 12 is a simplified diagram of a web page&apos; showing the configuration of uniformly formatted search results. Therefore, it can be seen that each search result list starts with a "model" string, and a "product description" is provided below it. In addition, "Price 1" and "Price 2" are to the right of "Product Description". 1280488 V. INSTRUCTIONS (23) Figure 13 provides a simplified diagram of the same screen used for these search results, showing the use of the vertical isolation to display the catalog of products. This vertical format can be categorized as the head, content, and trailer of the file. For example, the original design of the information infrastructure of the Internet, the training structure, the format of the product description of the online seller, and the expression of the technology are intended to be used by humans. This is understood in the use of query mechanisms and output standards, which are particularly suitable for direct human manipulation. Online sellers are in compliance with these rules because they facilitate online sales to human shoppers or buyers. Although it does not guarantee that making online stores easy for human navigation will make the AI software agent friendly to the owner, the online intelligent information comparison of the multi-language electronic data source of the system of the present invention is designed to achieve the benefits of these laws. . In accordance with the present invention, the construction of the package is implemented via inductive learning. The methodology learns the burden of the vendor by inferring the sample of the seller's web page. In the methodology of the present invention, the examples correspond to the vendor web page, the logo of the web page corresponds to its associated content, and the hypothesis corresponds to the parcel being built. Moreover, in accordance with the present invention, a parcel level that can be effectively learned, such as a HLRT pack level, is included. Also, in order to ensure that the methodology is well implemented, the noise allows the technology to be used when the training data exhibits a high level of noise. For example, a single-screen example of wu 'J j.0 · coin in Figure 15A is known. The smart price recognizer can distinguish between "price j" and "your price". The identified instance can be confirmed to mark the entire web page. Considering the other one of the item's identifier and price, it is confirmed that a labeling methodology is generated that contains a web page containing these attributes in pairs. 26 1280488 V. INSTRUCTIONS (24) In fact, the seller can create a sense of resolution by using a unified view of the type of product. For example, the seller presents an MD product information in the same format as the DVD product. By adopting the benefits of this regularity, each product hypothesis can be described in the same format.

The semantically resolved learner agent 18 in Figure 2 learns the package only from the specific domain of the example and attempts to apply this domain to all other domains (all other page categories within a completely different ontology project). The consistent format is organized and is categorized online on the 245 of the World Wide Web's remaining Internet-connected countries. Thus, the learner agent 18 maintains a full update with the semantics of Figure 2 of the present invention. The naming of the global product database is feasible, without having to write code for the SQL-compliant database, or manually typing each product into each domain for the Microsoft Acess database. Continuing with the sixth step, in step 210, the semantically resolved learner agent a generates a set of indicia to represent the content of the training web page. In another way, the methodology for indicating location values is to identify location information for training products in the training web page. Again, the semantics of Figure 2 distinguishes learner agent 18. Instant, rush, and automatically generate an indication containing the following position values: eKi&gt; ^ ^Κρ^1ίΡ&gt;Ν &lt;b:!·丨.e2 丨&gt ; ' &lt;b2 p e2,p&gt;&lt;b3.i,e3"&gt;,&lt;b3,p,e3,p&gt;

&lt;V, · e. 〉,&lt;b4.p, e4,p&gt;J For details, please see the right side of Figure 8. 27 ---- 1280488

V. INSTRUCTION STATEMENT (Μ) At step 220, the semantically-resolved learner agent 18 performs inductive learning using the relevant indicia for the trained training web page to output a set of possible candidate "vendor descriptions". Since the candidates are generated from specific training web pages with specific training materials, the candidates may not be invalid for these pages. ...: and if the candidate is valid throughout the vendor's website, cross-web authentication must be implemented to derive a generalized vendor description that will be valid throughout the vendor's website. At step 240, a vendor description authenticator (VDV) authenticates the possible candidate vendor descriptions for another set of training web pages (stepped in step 13 of Figure 3). If a vendor description is satisfactory, the learning process will stop (see Figure 3), otherwise the authentication process will continue to improve the vendor description selection by using the remaining candidates and through the loops of steps 230, 240 and 250. One of the criteria described by the vendor as "satisfactory" is that the candidates for each subsequent analyzed training web page are identical in number and character to the previous training web page. The training webpage analyzed on the right and subsequent is different from the training webpage of the candidate. Then another training webpage must be analyzed. Sections 10A and 10B provide examples of the left and right markers of the candidate vendor description for item description, price information, and the head and tail of the training web page. An example of "seller description" is provided in Figure 5, which includes identification tags for item descriptions, price information, and headers and trailers of training pages. Training materials for a particular vendor are preferably used by system administrator 26. As described in further detail herein, in order to add a particular vendor to the system, the vendor name of the corresponding training instance, the vendor URL, the URL of the proposed form, the domain data are provided and stored in the offline database 24, 28 1280488 V. Description of the invention (26) For example, it can be the Microsoft Acess database. The vendor name will be the base key for a record. Artificial packet input can be provided as an alternative. In order to provide a precise training case data set (this will instead greatly enhance the semantics to distinguish the accuracy and efficiency of the learner agent 13 in preparing itself for the instant, automatic, and rush to generate vendor descriptions) It is very important that the training case information prepared by the system administrator or other individuals can be used to understand the URL of the webpage and the name of the domain name to know the local language being used in any multilingual website. Learning the information type of the goal. This person does not need to know the code. Once the seller's information is provided, the administrator 26 can perform semantics for each vendor to resolve the learner agent processing. After the administrator has performed the semantics for the vendor to resolve the learner agent 18 once, it can execute as shown in steps 16 to 23

The graph shows any desired options as it walks through the interactive agent character learner interface (IACLI) screen through the learning process. Finally, the "seller description" result set retrieved from the returned training page will be stored in an offline repository 24 such as the Microsoft Acess database. In order to delete/remove a particular vendor, the administrator can delete the record directly from the "seller queue" or "seller description queue". For modification/editing, the administrator can modify this record from the "Seller's List" or "Seller's Description" column in the folder. In a small container, the semantic discriminative learner agent 18 of the present invention produces a unique vendor description of a particular vendor. The seller describes a set of generalized rules for how a vendor organizes its product information in a particular format. Thus, the input to the package building system of the present invention is essentially a sample of the behavior of the package to be learned. Under this shape, the package construction becomes a reconstruction of the package according to the package behavior.

29 1280488 V. Process of invention (27). The methodology of semantically resolved Learner Agent (SRLA) 18 is extracted in Figure 14 in a simplified example. In step 1, two pieces of information are fed into the system for parcel induction: (1) the URL of the seller's website (eg http: / / WW - W · 800 ) · C Ο spot (2 V network is also ice, its Including certain training examples for a specific domain. For example, a domain description can be "electronics, and the record for this domain can be "Sony ΗΜ3 81MD", which is the model used to fill in the vendor's search form. In steps 2 and 3, the semantically resolved learner agent 18 automatically uses the URL and the domain/model of the training examples via the Internet to the vendor website. For a specific example, the semantically distinguishes the learner agent. The device will proceed to the www.800 page according to the URL provided in step 1. Then it will fill in the necessary product information with the relevant search form (ie domain description "electronic product" and "HM381MD"). Finally It will "propose" the search form to request a search and wait for a response. Referring to step 4, a result web page is sent back according to the search criteria. This result can be a successful result web page with a related product description, or a failed web page. It will be noticed, being The content of the returned web page is the HTML code, the item description, the item price, and the location of the information for the HTML code. In steps 5 and 6, the search result web page is sent back to the semantically resolved learner agent via the Internet. For analysis. The analysis performed in step 7 is referred to as "packaging induction", where the web page is generalized into a set of permutations and formatting rules, and the vendor logically abides by these rules to present its vendor description. Separately distinguishing the buyer agent 20 from these rules can be used to buy product information from the same vendor when the buyer/user searches for certain product information from the same domain of the vendor website. It is understood that semantically distinguishing learner agent processes in accordance with the present invention

It will be implemented for each seller, and a seller description will be derived from it. Due to the information tagging practices used in the present invention, the vendor description can be obtained in any language from any vendor website. Succinctly speaking, although the language presented to the user may be a particular local character string, the desired information may be identified as The marked code based on the language remains the same regardless of the local character string of the language. In other words, the information described by the seller will be obtained from the web page in the local language used by the seller's website. It does not require any translation of the local language into a standard language. In addition, because each vendor's website is identified as a candidate that is not written on the basis of the programming language used for the web page, subsequent searches may be completed without the need to use different programming languages in the web page being searched. The buyer agent 20 is distinguished from the multi-domain (product category) and is searched independently of any programming language. Referring to Figure 15C, the flow circle 300 shows the semantic resolution of the buyer agent 20 of the present invention. In the step 310, the semantically distinguishing buyer agent 20 receives a request from the buyer/user of Fig. 2, which requires a price comparison of the product. In step 310, the buyer agent 20 is semantically distinguished to communicate. Use

The Active X component also preferably establishes a link to the user/buyer 12. The user 12 must provide at least one parameter which may, for example, include the desired product name, the range of desired prices, the online vendor or storage criteria of the target. Step 312 checks the memory of the desired information contained in the identified vendor website. Whether there is a "hit" in the body or cache memory; if so, semantically distinguishing the buyer

31 1280488 V. INSTRUCTIONS (29) The agent will proceed to step 370 to store the extracted target information. The semantically distinguishing buyer agent will then proceed to step 380 to generate a resulting web page from the target information in the HTML, and in step 390, the buyer agent will be visually identified to display the resulting web page to the online human user/buyer. If a "hit" is not found at step 312, step 320 will summon the semantics to distinguish the buyer agent 20 to use the input parameters to retrieve the vendor descriptions from the offline database 24. These "seller descriptions" are semantically distinguished learner agents 18 that were previously defined when the vendor described the learning process. In step 3 3 0 ', we know that the shell agent 20 will combine the new request of the user to access the different online sellers identified in the "seller description queue". The new request of the composed user will be based on the parameters given by the user and the information in the vendor description. Preferably, if there is one of the requested online sellers (e.g., product model request), it will have a new request to be semantically distinguished from the buyer agent 20. The buyer's agent 20 is determined to use the vendor description to obtain real-time price information from the seller's website. The semantically distinguishing buyer agent 2 uses the vendor's URL and the vendor name included in the information describing the vendor description to navigate to the vendor website', which is also included in the vendor description as the vendor's search form URL. In step 340, after navigating to the seller's website, the semantically distinguishing buyer agent 20 "fills the seller search table "virtually" according to the user's new request and "virtually" presses the enter key to present it. This is done after each identified online seller. As described above, the seller description in the offline database 24 includes a block, which &amp; the seller's search form URL ', such as "http://www.oiiliiieshop.32 1280488

V. Invention Description (30) com/search.asp?item=”. Separately, the buyer agent 20 uses the user's input parameters and the search form URL to combine new HTTP requests for each identified online vendor. For example, If the user wants to buy a "hard disk", the new request to be semantically distinguished by the buyer agent 20 combination will be as follows: "http://www.onlineshop.com/search.asp?item=harddisk,'' The agent 20 will send the HTTP request to the online vendor as if the user had made the request. If there are N identified vendors, the semantically distinguishing buyer agent 20 will launch N execution lines to fill in the search for N. The semantically distinguishing buyer agent 20 is preferably parallel to the searchable metrics of each online vendor, populating it and proposing a search request. In the semantic sense, the buyer agent 20 will be tentatively specified or used. Waiting for a response from the online vendor within the defined pause. If a pause occurs, the semantically distinguishing buyer agent proceeds to step 370; otherwise it proceeds to steps 358 and 360 to further process the received search result data. During the pause period, the semantically distinguishing buyer agent 20 collects responses to search requests from different online vendors. In step 358, the semantically distinguishing buyer agent 20 receives the search result response from the online seller and stores it in the servo. The memory or memory within the cache 22 is retrieved from the received response at step 360. The semantically distinguishing buyer agent 20 uses the vendor stored in the vendor description 28 or offline database 24 Describe the information in the queue to extract the desired information. For example, the vendor description includes the identification of the code for the left and right packets. First, the semantically distinguishing buyer agent (SRBA) 20 will use the left package information. In order to locate the effective information 33 1280488 5, the invention description (31) in the response page, the correct location of the target data (such as the information in the seller description column is defined) will be stored and stored in the memory. (Recall that in the learning process of Figures 3 and 6, the learner agent 18 has been semantically distinguished. The learner is the information in the seller description.) The extraction of the target data will be repeated. Until the end of the page. For example, during the extraction process, the product description and product price will be extracted. It will be understood that the information contained in the vendor description defined by the semantic learner agent is independent of the domain independent but more Language-related. For example, 'Assume that the online buyer or buyer's platform uses the Windows 98 operating system and executes the language version "B, (or preferably its platform is the English version of Wmdows 2000 professi〇nal and / or its platform) Install the personal web server version "B"), Microsoft Internet Explorer will urge him to download the "B" language display software after logging in to the portal of the invention. In the online buyer user A 'type a product model in the local character of the language 'B' as the keyword in the text box provided by the entry of the present invention, the semantics in Fig. 2 distinguishes the buyer agent 20 The data extraction will be performed using the pre-described example data (slightly earlier than the instant package induction learning in the seller description of the advocate), and the product model represented by the language "B" contains the local character string. These vendors Also described in the pre-defined data structure is stored in the vendor description column in the offline database 24 (preferably the Microsoft Access database). The data extraction involves hitting the r" which contains the previous search results. The price, description, and associated product information network temporary search uses the correct local character word for the product model represented by the language "B," as entered by the user in step 312 as shown in Section 15C. The string resides in the server 22 34 1280488 V. The invention (32) is stored in the memory or cache memory. Since the character string of the language "B" is a specific local language, any is The "hit" found will be used for the identified vendor website using the local language "B" and has the character string represented by the language "b". Recall that in step 7 of Figure 14, the vendor is described in the learning process. In the second diagram, the learner agent 18 has been learned by the online vendor and stored in the offline database 24 (preferably Microsoft). Access database) or database server 22 (preferably Microsoft compliant SQL database server). Since it is inefficient to distinguish the buyer agent 20 from the request of the online buyer user to retrieve the information of the seller description, the seller description is fortunately only in the sense of the other. The search-conform-extraction of the desired set is requested by the offline database 24 or the server 22. The vendor description will then be stored in memory or cache memory for more instant insertion and for subsequent requests by the same or new online user to distinguish the buyer agent 20 later. The vendor description in memory or cache memory is preferably automatically updated daily. In other words, the semantically distinguishing buyer agent (SRBA) 20 can use the material in the vendor description to locate the target data in different domains and in different languages. This is because, for a particular vendor, although the language will change, the underlying code corresponding to the target information will not. Since the three r-format rules pre-exceed most seller websites, such as B to C, C to B, C to C online stores, etc., different domains on the seller's website will consistently use the same format and base code to present the target. Information such as project descriptions and prices. 1280488

V. INSTRUCTIONS (33) So in terms of the search response that the parent is sent back, the semantically distinguishing buyer agent 20 will use the vendor description to implement the data extraction. If the time is up, the semantically distinguishing buyer agent 20 will proceed to step 37 of FIG. 15C. In step 37, the semantically distinguishing buyer agent 20 will extract data from different online sellers according to user-defined ranking criteria. sequence. If the user does not define a sorting criterion, the preset value will be the price of the product. Alternatively, the ranking criterion is to identify the best price found and present only the information from the seller to the user/buyer at the best price (in other words, other sorting criteria can be used). After the sorting is completed, the semantically distinguishing buyer agent 20 will proceed to step 380. In step 380, the semantically distinguishing buyer agent 20 combines the HTML web pages based on the information filtered and sorted by step 370. In step 390, the semantically distinguishing buyer agent 20 responds to the user request by presenting the combined HTML web page as a "result" web page using the previously created ActiveX element. If the time has not expired in step 350, the semantic resolution of the buyer agent 20 will proceed to step 358. In step 358, semantically distinguishing the buyer agent 20 from storing the data of the search result web page in the memory or the cache memory to instantly respond to the new request of the same user/buyer or the request of the new user/buyer use. Following step 358, the semantically distinguishing buyer agent 20 will proceed to step 360 where it extracts the target information from the search results web page, sorts the results in step 370 according to the user-defined ranking criteria, and is shackled by different online vendors. The extracted data is then combined, and the HTML web page is combined in step 380 based on the data filtered and sorted by step 370, and finally the previously created ActiveX element is used in step 390 to the user/buyer back 36 1280488

V. INSTRUCTIONS (34) The default language of the buyer agent 20 should be determined in English as English. With the pre-set, semantically distinguishing the buyer agent 20 will go to all the sellers when it receives the user's request. When the response is sent back, the semantically distinguishing buyer agent will use the vendor description that has been semantically distinguished by the learner agent 丨8 to filter and invalidate the result. In another embodiment of the present invention, the vendor can be categorized to the user's premises so that the user 12 can select "Advanced Search" to search for the group to which the seller is classified. The methodology utilized by the present invention is inherently multilingual. When the semantic learner agent 18 learns a vendor website, this learning can be implemented in the local language of the website. The results of the descriptions that have been captured and used to portray the site for the vendor will be the local language of the site. Thus, when the online user/buyer submits a request in a specific local language at step 310 of Fig. 15C, the user is determined to use the correct character string typed by the user in step 312. In the memory boat or cache memory. Since the character string will be in a particular local language, any "hits" found will be used by the vendor website that uses the same local language and has the same character string identified. In the case of this methodology, it will be appreciated that there is no need for a "translation" step in accordance with the present invention to transform a search request made in one language into a "standard" language. By using the local language of the search request, errors and ambiguities caused by the translation steps are avoided. A development tool for a database server (preferably a Microsoft compliant SQL database) that is used in a preferred embodiment of the system of the present invention is established. 37 1280488 V. Description of the Invention (35) The computer program module is Standard, and the present invention may use an Oracle database such as Redwood Shores, Inc., which supports ODBC, a SQL database server from Sybase, Inc., Emeryville, California, and other related databases. As described above, simultaneous search for multiple execution line processing is important to the preferred embodiment of the present invention. In this regard, the use of the Windows NT 40 platform (a product of Microsoft Corporation) provides this multi-line processing capability. The compilation and preparation of the training materials in the "Seller's List" 27 and the information in the "Description of Sellers" 28 will now be operated in more detail in the light of the 16th to 23rd. Figure 16 is a real picture of the Interactive Agent Character Learner Interface (IACLI), which can be used to obtain training information for use in the present invention. Using the screen corresponding to the "Join Seller" logo shown in Figure 18, the login point for the information obtained by the system administrator after viewing the seller's website "lcache.com" is provided. Thus an example is provided in the 17th, where this information has been typed in for the vendor "lcache.com". This information includes those shown below in the right column: i vendor name 1 cache.com seller URL http://www.cache.com ί seller search form http://st4.yahoo.com/cgi- URL bin /nsearch?catalog=lcache&amp;query= Learning Domain Dvd Training Example i) DVD Virtual Notebook with I-Glasses ii) JVC XV523GD Dolby Digital DVD Player iii) Pioneer DVL-919-Combined LC/DVD/ The CD player then stores the information in the seller's queue 27 as offline data 38 1280488. 5. Description of the invention (36) Training materials in the library 24. It must be noted that the training case that has been typed is a list of specific products that will be searched for during the training process with the designated identified seller website that will be acquired on the training web page. The "Seller Description" will then be "learned" from the returned training pages. The information can then be displayed in the learning as shown in Figure 18 corresponding to the "Seller Information" logo. The "Seller Information" screen interface (see Figure 19) provides a "search" function for the vendor name. By typing a vendor name and pressing the "Search" button, the vendor information of the typed seller is retrieved and displayed by the offline database 24. In this "Seller Information" screen, please note that "Package" - "Head", "Tail", "Item Left Mark", "Item Right Mark", "Price Left Mark", "Price Right Mark" are empty. of. Section 20 provides a single screen of the learning interface for displaying training examples that were previously typed for a particular vendor. This screen is displayed in response to the opening of a file called "Training Data". For the "Seller Information" screen on the learner interface, there is a search function screen provided for "Training Information". In order to use the training data search function, the system administrator types the vendor name and presses the "Start" button. This triggers the "training data" column that was previously typed for a particular vendor to be displayed as shown. It should be noted that the "Training Information" interface provides other functions such as "join" (additional examples), "delete" (delete a training case), "edit" (edit training example), save (the list of examples) Save to its current state), Cancel (cancel the typed change). Referring now to Figure 21, a single screen of the learner interface is displayed in response to the opening of a file called "Seller Description". This interface begins the process of "learning" a vendor description and provides the system administrator with the vendor "39 1280488 5, invention description (37)" (the training material has been typed) or the "some" vendor. (The name has been typed by the administrator in the box provided) to learn the choice. In Fig. 22, the "learning one" is selected and the name of the seller is "lcache.com". When the system is connected to the World Wide Web, the "Learning Now" button will be depressed to launch the semantic learner agent (SRLA) 18 to "learn" the vendor information about the particular vendor "lcache.com". The method is to use the training example determined in the training example of this seller on the website. At the end of the learning/training process, the results of the training or study that have been sent back by the website are displayed on the learner interface as shown in Figure 23. Again, to display this information, the system administrator can use the search function on the vendor information screen of Figure 19 to type in the vendor name ("lcache.com" in this case) and press the search button. Figure 23 shows the learning result of the seller “1 cache.com” in an instant. It should be noted that the "package" has now been completed. Similarly, the "header" of the web page is displayed to occur at the value "5230". This value "5230" specifies the information for the line, character position or other location. "Tail" indicates that the position of the item is as follows: "D &gt;&lt; /TD &gt;&lt; TD &gt;&lt;/TD>&lt;/TR&gt;&lt; /,, on the item For the description of the information, the left mark is defined as the following string: "G SRC=/Lmg/trans+ lxl.gif BORDER_OWID···,, the right mark of the item description is set as a string: "&lt;/b&gt; ". The left mark of the price is set as a string: "&lt;/b>&lt;&gt;/A&gt;&lt;/TD&gt;&lt;TD ALIGN = rightxFON··.,, finally, the price right mark is set to Such as the string "&lt;/T&gt;". 40 1280488

V. INSTRUCTIONS (38) Although the left subtag of the item in item 23 and the character substring of the right mark of the price are cut off due to the static display state of the learner interface, it will be understood to be semantically distinguished by the learner agent 18 All of the characters of the left-tagged string will be stored in the vendor description 28, preferably in the Microsoft Access repository' and will be used semantically to distinguish the buyer agent 20 later. Now, the methodology that is the basis for semantically distinguishing learner agents 18 will be described in more detail with the degree of "concept proof." The package induction problem is shown in Figure 24 as a simple model of information extraction. As shown in Figure 24, web page p is the web page containing the desired information. p is used as a string of certain letters. Typically, the letter is an ASCII character set and the web page is an HTML file. For example, Figure 7, which was operated earlier, is a very simple web page obtained from a seller's website. In the case of "marking terms", this page will be referred to as PPC (the web page of the product catalog). Note that the methodology of the present invention is initiated or activated using HTML but is not dependent on this. For example, such web pages may be natural language text or may conform to XML standards. A standard relational data model was adopted. Related to each product record is a bipartite attribute · project and price. Here, "item" represents the product name or model number, and "price" represents the price of the product. A "tuple" is a two-string vector &lt;Ai, Ap&gt;. The string Ai is the value of the "item" attribute, and the string Ap is the value of the "price" attribute, and the attribute represents the line of the relational model, and the "tuple" represents the column. Therefore, as shown in Figure 8, the example of the catalogue page of the seventh page contains four "yuans".

1280488 V. Inventions (39) Group, the first of which is &lt; HM 381 MD,, ‘399.95’ &gt;. The content of the web page is a group of "tuples, for example, the literal string token is appropriate, but because the webpage has an unbounded length", the clearer and concise presentation of the webpage content is used instead. The logo of the web page is used to accurately present the content of the web page for a set of indicators within the web page, rather than to explicitly list the attributes. For example, a simple product catalog page (PPC) is displayed with a "mark" Lpc. The right side of Figure 8 is rotten. "Mark" means that the simple product catalog contains four "tuples", where each "tuple" consists of item and price values. A pair of integers represents each value. Considering the first pair &lt; 174, 1 80&gt;, these integers indicate that the attribute of the first tuple is a substring between positions 174 and 180, i.e., the string 'HM381MD,. The check of the character string on the right side of the figure reveals that these integers correspond to the character position, starting with "&lt;HTML&gt;," in the first line. Similarly, the last pair of integers in the fourth "tuple one" &lt;356,361&gt; indicates the last attribute, and the price appears between 356 and 361, that is, the string '399.95,. More generally, 'page P The content can be labeled [representative. 丨p&gt;, L== ek&lt;i&gt; » &lt;bk p ek p&gt; [ , bm, 丨,, &gt;, &lt;bm,p em p&gt;j has only a single "yuan" For the webpage of the group, the result is the following: eu&gt;, &lt;bl p, ei p&gt;&gt;} 42 1280488 V. Description of Invention (40) The label L encodes the content of the web page P. This page contains the "tuples" of |L|&gt;, each of which has two attributes, items and prices. The integer l&lt;m&lt;|LI sets the "tuple" in the web page. Each pair &lt;bm〆em丨&gt; encodes an item value, and each pair &lt;bm,p,6„1,|)&gt; encodes a price value. The value bm i is an item in P The starting indicator of the mth "tuple", the value em i, is the ending indicator of the item value in the mth "tuple". Similarly, the value bmp is the starting index of the mth "tuple" of a price in P, and the value is the ending index of the price value in the mth "tuple". Thus, the item attribute of the πι tuple Appears between <IVi, 6„^&gt;, the price attribute of the mth tuple appears in &lt;bm,p,em,p&gt;. So the pair in the 8th is &lt;b2.,,e2,,&gt ; = &lt;229,234&gt; The example encodes the second (item) attribute of the second "tuple" in the simple product catalog of the web page of the seventh page. As shown above, a packet W is a function indicated by a web page to a mark; the symbol W(P) = L indicates that the result of exciting the packet w on the web page P is the mark L. At this level of abstraction, the package is only a random program. A parcel is a set of packages. As will be seen later herein, the package used in the present invention is referred to as the HLRT package level. The previous explanation of the terms and formulas used to describe the methodology of the present invention will now further illustrate how the learner learns the vendor's catalog page. Intuitively, the input to the learning system of the present invention is a sample of the catalog page and its associated "mark". At this point, it is assumed that the "marker_ has been determined and given. Further efforts to generate the method for the sample web page will be provided later here. The output is a parcel. 43 1280488 V. DESCRIPTION OF THE INVENTION (41) Ideally, W will output appropriate indications for all sample web pages. In general, such guarantees cannot be made, so (in the spirit of inductive learning), it is required to generate correctness for a certain set of training cases. In the case of problem solving, the package induction problem (for a specific level is as follows: Input: a set of training examples of ε = {..., <卩!!, Ln&gt;,··}}, where Pn is one The web page, and each Ln is an indication; Output: a parcel WeW, so that for each &lt;Pn, Ln>€ ε, W(Pn)=Ln 〇 as explained earlier here, the pcwrapHLPT program describes a word Header design uses the head tag, left tag, right tag, and tail tag of the vendor's catalog to extract relevant product information and its price. This head is left and right (HLRT) package level for this word. One of the methods of head formalization. Figure 25 is established The "execHLRT" program is a generalization of pcwaHLRT, which allows these tags to be arbitrary strings instead of the specific values such as "&lt;B", &lt;/B&gt;" used in the previous implementation of pcwrapHLPT. The mark in this example is the entire HTML mark, and the method of the present invention is not limited to operating with HTML tags. Moreover, the text may not be HTML at all. Therefore, the amount symbol "$" may be a price such as "S399.95". The effective left mark. This execHLRT subroutine determines how the HLRT package works. Earlier it was pointed out that W(P) is the indication of the result of stimulating the package of the web page P. This sub-executive execHLRT is used to determine W (P) The procedure and the case where W and P are used when W is a HLRT package. 44 1280488 V. Description of invention (42) li and η represent the left and right marks of the item attribute, and % and % represent the price attribute. The left and right marks, and 11 and 1 indicate the head and tail of the page. (Note that 'h is the line number instead of the string. For example, if h=i〇〇, then the page is 100 before the head, semantically distinguishing the buyer agent 2 skips this when it searches for a product Some lines.) For example, if execHLRT is h=7,1 &lt;Β&gt;" '&lt;/Β&gt;', 1ρ=“&lt;ι&gt;”, “〇,, and t==: “&lt;/ TABLE:&gt;" is activated, then execHLRT acts like pc^vrapHLRT. More generally, any HLRT package used by a seller's website is equivalent to a vector of (h,1,,Γι,1ρ,Γρ,t), And any such vector can be interpreted as a hlrt package. Under this equivalent, the notation (h, η, 1ρ, Γρ, t) is used as an abbreviation for the HLRT package obtained by partially evaluating the execHLRT with the certain mark. Since the HLRT package is only a vector (h, 1丨, η, Γρ, t), the HLRT packages of Figures 7 and 8 are summarized in a set of pages and their marked £ = !...CP, ( Pn, Ln), ···} is based on one of the four markers. More precisely, the following constraint satisfaction problem (CSP) is to be solved: Variables: Header of page P: h End of page P Marker: t Left attribute of item attribute: li Right attribute of item attribute: η Price attribute Left mark · lp Right attribute of price attribute: rp 45 1280488 V. Description of invention (43) Definition field: Each tag is any string restriction except the header tag: just &lt;Pn,Ln&gt; €ε,iT( Pn) = Ln, where HLRT packet 袠 W = (h, I;, η, lp, rp, t) Now learnHLRT will be described, which is proposed for the above limitation satisfaction problem. It is important to note that the domain of the tag variable is strictly limited by the example ε. At a minimum, the tags will be substrings of the page instance. Of course, we can do better. Based on a single case (Ppc, Lpc), it can be seen (the right mark of the price attribute) must be one of the words "&lt;/l&gt;&lt;/TD&gt;&lt;/TR&gt;, where "xlT Represents a new line character. The "word prefix" is a combination of characters in a string, starting from the rightmost character of the string; for example, "&gt;", "D&gt;,,, "TR&gt;", etc. Note that if rp is not this word At the beginning of the string, each package with this tag fails at least when the fourth "tuple" of the Ppc extracts "399.95" as the code attribute. Therefore, the candidate for rp is all the prefixes of "&lt;/b&lt;/TD&gt;&lt;/TR&gt;". These candidates are shown in Figure 10A. In detail, the candidate tags for the simple catalog page are generated as follows: Consider the left mark lp of the price attribute. Recall the segment before the price in Figure 7 "HM381 MDx/Bx/TDxTDxl&gt;", "MD2070&lt;/BX/TD&gt;&lt;TDxl>" and so on. It is assumed that under these segments, lp must be the suffix of "&lt;/B&gt;&lt;/TD&gt;&lt;TD&gt;&lt;l&gt;". Therefore, the candidate for Ip is the 16 non-empty suffixes of this string. This candidate can be seen in Section 10A. The so-called "suffix", meaning the combination of the characters of a string, begins with the leftmost character of the string, such as " &lt;", 46 1280488

V. Description of Invention (44) "&lt;/,,, "&lt;/B", "&lt;/B&gt;" etc. The 0 mark li is the most complicated because the string before the first attribute is in the previous The first attribute and the last attribute of the "tuple" appear between the first "tuple" at the beginning of the page. In this example, the string considered is " &lt;TR&gt;&lt;TD&gt;&lt;B&gt;, and 44 &lt;/l&gt;&lt;/TD&gt;&lt;/TR&gt;&lt;TR&gt;^&lt;TD&gt;&lt;B&gt;". It is clear that li is the suffix of this string. Therefore, the candidate for li can be generated by enumerating the suffix of this segment. To generalize this effort, it is summarized that given the set of examples and the candSi(i,p,s) written, the candidates for the marks li and lp appear in the project attributes or in each case by enumeration. The suffix of the shortest string to the left of each instance of the price attribute is generated (as mentioned in the previous paragraph, the project attribute is somewhat special. The shortest between adjacent tuples or before the first tuple The suffix of the string must be enumerated). For example, if ε = {(Ppc, Lpc)丨, then: candSj(i,p,E) = {&lt;/1&gt;&lt;/TD&gt;&lt;/TR&gt;nI^ &lt;TR&gt;&lt;TD&gt;&lt;B&gt; '...} cands丨(ί,ρ,ε)= {&lt;/B&gt;&lt;/TD>&lt;TD&gt;&lt;l&gt;, candidates for the right mark are similarly generated, sentence/have Two things are different. First, the string under consideration occurs to the right of the appropriate attribute (not to the left). Second, r, and rp must be the prefix (not the end of the string) of these strings. For example, in the simple product catalog example, the tag must be the prefix of the string " &lt;/B&gt;&lt;/TD>&lt;TD&gt;&lt;l&gt;, and % must be "&lt;1&gt;&lt;/TD&gt ; &lt;TR&gt;,, and &lt;/l&gt;&lt;/TD&gt;&lt;TR&gt; ^ &lt;TR&gt;&lt;TD&gt;&lt;B&gt;, both of which are prefixes. In particular, the set of examples is given. The right mark of the cands Jk, s) is written. 47 1280488 5. The description of the invention (45) by enumerating the prefix of the shortest string occurring on the right side of each instance of the attribute k in each case. Was produced. (Like the above, li is a special case.) Similarly, rp is a special case. The shortest string that occurs between adjacent "tuples" or the last "tuple" is listed. For example, candsr(i,p,e)= {&lt;/B&gt;&lt;/TD&gt;&lt;TR&gt;&lt;l&gt;,...} candsr(i,p,e) = {&lt;/1&gt;&lt;/TD&gt;&lt;TR&gt;vI^&lt;TR&gt;&lt;TD&gt;&lt;B&gt; '...} A similar analysis applies to the head and tail tags. The "header" is the prefix of the page before the first item attribute appears. Note that the "header" is presented here as a string. When a package is actually implemented, in order to improve the performance of the present invention, it is preferred that the "head" be presented as an integer so that the human shopper or buyer can use the package to find product information without having to look at the content. Quickly jump over the head of the page. To convert a header to an integer, just find the number of rows that the header string spans. The mark identifying the "tail" is very similar to the right marks li and Lp. These tails are the suffix of the string after the last price attribute of the page. Cands(head,E)={&lt;HTML&gt;&gt;|^&lt;nTLE&gt;&quot;.PRIEC($US)&lt;/rH&gt;&lt;/rR&gt; ,...} cands(head,£)= {&lt ;/l&gt;&lt;/TD&gt;&lt;/TR&gt;&gt;||&lt;/TABLE&gt;...&lt;HTML&gt; » Simple to learn these two tags under each of these tags The virtual program of the method is provided on page 26. Since the module is executed at a time proportional to the number of candidates for each tag, and since each tag has many candidates, the execution time will be 48 1280488. V. The invention (46) is very slow. A more efficient process can be achieved by observing the mutually independent markers Ip, rp. Moreover, a candidate is not associated with any other tag as to whether the particular store is valid. For example, it can assess whether “/B” is satisfactory to the factory without any reasonable arrangement of any other markings. To see that this independence is properly established, you must recall the execHRLT order. At each point of its execution, execHRLT searches for its input page P for any of the tokens ^, Ιρ’Γρ. If any of these searches fail in determining the correct position in P, the indication of the output by execHRLT will be incorrect. However, whether these searches send back the correct answer depends only on the mark of the consideration and the example of the page, rather than other marks. Alternatively, once a particular candidate (L, Ip, Γρ) is selected for some of the markers, the candidate will not be invalid regardless of whether the candidate is selected by other markers. The change of this claim also has an intuitive meaning; if a candidate is invalid, it cannot be repaired regardless of how carefully the candidate is selected by other markers. Note that this independence is guaranteed: this is not just an inspiration to promote learning. The importance of this observation is that the three markers Γ, 1ρ, Γρ can be learned in isolation. In the virtual code, it can be learned as follows: 1 - Generate candidate groups: 2. Select a valid candidate for each tag. This methodology is much faster than the procedure of Section 26; it is performed at a time proportional to the sum of the number of candidates for each marker (not the product). However, it has also been observed that not all markers are independent of each other. 49. INSTRUCTIONS (47) In contrast, in the case of the marks h, t and li, a particular character string is valid for one of the two markers depending on the choice of the other two. For example, is &lt;®&gt; effective? The answer depends on the choice of h and t. If h= " &lt;HTML&gt;", since execHLRT does not skip the irrelevant boldface "&lt;B&gt;A Simple Product directory&lt;/B&gt;", then "&lt;b&gt;,, η will not be A valid tag. Conversely, if h= “&lt;/TH&gt;&lt;/TR&gt;, then li= will not be a problem. Similarly, li interacts with t: if t:= “&lt;/HTML&gt; ”, 1ι= “&lt;B&gt;” is unacceptable, but if t= “&lt;/TABLE&gt;”, it is Again. As a result, the two candidates h, t and li must be considered jointly. Therefore, all combinations of candidates for h, t and li are enumerated, and valid ones are selected. This improved second step involves characterizing the conditional conditions that effectively mark the candidate. First consider the marks η and Γρ. After the method has determined the beginning of some instances of this property, the method attempts to determine the end of this instance of the property. Therefore, the flag "or candidate" u, must satisfy the second limit: Limit Cl: "u" must be a substring of any instance of any of the attributes of any web page instance. The prefix of the text that must appear immediately after each instance of this attribute in each web page instance. If these restrictions are flagged by ri or rp - the candidate "u" is violated, then each packet - at least - of the example e is failed. If the limit is violated, the attribute k will be too short; if C2 is violated, it will be too long. Consolidation and &amp;, what is said, if a candidate "u" is to be effective 1 1280488 5, invention description ( 48) When marking the value of η or Γρ, these conditions must be true for the case of a certain set of ε. These conditions will be called valid(u,r,£). It is seen that if and only candidates When "u" marks η and Γρ for the example set ε to satisfy the limits C1 and C2, validr(u, γ, ε) holds. Back to this example, if the vaUdr check is applied to the candidate generated by candsr , you will find: For the right tag of the project attribute: validr(&lt;/B&gt;&lt;/TD&gt;&lt;l&gt;,i,ε)= true for the right tag of the price attribute: validr(&lt;/ l&gt;&lt;/TD>&lt;TR&gt;do&lt;TR&gt;TD&lt;B&gt;, ρ, ε)=false the execHLRT program search mark lp. The candidate of the mark ιρ "u,, must satisfy the second limit: limit C3 : "u" must be an appropriate end of the text that appears immediately before each instance of attribute k of each page instance. If this restriction is violated, each package will not Meaning ε. At least, the opening indicator bm, ρ (calculated by execHLRT) will be incorrect - it will be less than or greater than the correct value, or undefined, depending on "u, how to violate this limit. If some candidate "u" is based on the example set ε as the effective value of the mark lp, the conditions discussed must be established. These conditions are called v^d1(U,l,〇. It is seen that if Only if the candidate "u," marks ip for C to satisfy the limit C3, then validi (u, 丨, 〇 holds. Go back to the simple product catalog example Ppc, which is seen: validj (&lt;/B&gt;&lt;/TD&gt;&lt;l&gt;,p,ε)=true, 51 1280488

52 1280488

V. INSTRUCTIONS (5〇) (1) The module heuristic search methodology is used to describe the composition (LabelOracle) as described in Figure 15C. This is called a resolver: one of them is the project 'resolver and the other is a smart price resolver. (2) It is inefficient to retrieve the vendor description data each time the buyer agent 20 receives a request from an online human buyer or user, if such a description is semantically distinguishing the buyer agent 20 first. The search, match, and extraction of the desired request will be retrieved only by the database (preferably a Microsoft Access database) or the database server 22 obeying SQL. The vendor description will then be stored in memory or cache memory for use at other times when the later request is resolved by the buyer agent 20. (3) The description of the vendor in the memory or cache memory will be automatically updated once a day. (4) The system of the present invention creates multiple execution lines and simultaneously deletes several semantically separated learner agents to access various designated online sellers through the World Wide Web. The use of this multiple execution line methodology is preferably built on DCOM provided by Microsoft Corporation. Each semantically distinguishes the buyer agent 20 from the product information provided by the human buyer or user and virtually presses "type" to intelligently fill the seller's search form. (5) On the other hand, the semantics of the present invention distinguishes the buyer agent 20 by speeding up the vendor response time and, if the multi-execution line is used to distribute the returned search result web page to the separate memory location, Shoppers/Buyers buy online for the entire process of global information online on the big internet traffic. Obtaining a training webpage involves an example query of the seller's website. Example 53 1280488 V. INSTRUCTIONS (51) If 'FIG. 12 shows the appearance of an example web page by querying a website such as httD://www.800.com. The ruler, called LabelOracle, consists of a modular heuristic search methodology that will now be described in more detail. A resolver finds an instance of a particular attribute on a web page. For example, given the example web page of Figure 12, a project resolver will find all the "projects" included in the web page, such as the products "HM381MD" and "MD203,". A resolver must be smart enough to remove this. For example, in the case of Figure 12 again, the intelligent price resolver can distinguish between 1 price 1" and "price 2", which are, for example, "price" and "your price". These resolver instances are then confirmed to mark the entire web page. For example, to give the "item" discriminator and another "price" discriminator, it is confirmed that a LabelOracle is generated, which marks the web pages of these attributes. If the project attribute knows all the items in advance, distinguishing the "projects" is a simple model matching problem. However, this requires a large queue of project names/models, which is not feasible. In addition, maintaining such a large price database is costly. Therefore, it is guaranteed that the name/model number of this project is completed and updated to be unpractical. Fortunately, sellers create a sense of identity by using a unified viewing attempt for all products. For example, the seller presents small disc (MD) product information in the same format as the DVD product. By adopting the benefits of this rule, it is assumed that each product is described in the same format. The present invention learns only one package from the example of a specific domain, and attempts to put this domain on the Internet to be organized in a globally consistent format. Mandarin 54 1280488

V. All other domains of the invention (52). In the preferred embodiment, the training examples originate solely from a domain, such as the MD domain of the vendor website. This results in the formation of a project discriminator that only has to distinguish the specific domain of the product such as MD. In this mode, it is possible to maintain a fully updated nomenclature for a particular domain of the project name.

The present invention recognizes "price" by stimulating a module-inspired search. For example, the price always follows the amount sign ($) and the price is usually a floating point number. If more than one price is found, an item such as "your price", "our price", "price", "original price" and so on will be extracted. The mechanism for semantically distinguishing how the buyer agent 20 operates is shown in Figures 14 and 15A-15C, as briefly described earlier. The control process consists of the eight steps marked in the figure: Step 1: When the user determines that a specific product or service is required, 'substituting one-to-one to manually search for product information and prices to browse different global languages on the global information network. At the seller's website, the present invention provides an entry for the request for product information via the interactive agent character user interface (IACGUI), which is generally known as an interactive agent character shopper/buyer. The interface is typed once for the same purpose, but with better, faster and more reliable results. The product description for this result is stored in the membership variable m_ProdDesc of SRBA 20. The search also enables users to customize how the agent works through the "Advanced Search" function, which provides optional parameters such as 55 1280488

V. Description of the invention (53) Select the seller, time limit (restriction), price range, any manufacturer, keyword, etc. Step 2: Do it again. For another example, the online browser or user platform uses the Windows 98 operating system and executes in the language version "B" (or preferably its platform executes the Windows 2000 in English version and/or its platform is installed in the personal web language version "B") Microsoft Internet Explorer will urge him to download the language "B" display software after logging into the platform of the invention. When the online buyer or user ''A' enters a product type in the local language "B" as the keyword in the text box provided by the entry of the present invention, the second meaning is to distinguish the buyer agent 20 The data extraction will be performed using the pre-described example data (obtained earlier in the vendor description after the instant package induction learning), and it contains the local character string of § § B. These vendor descriptions are stored in the database 24 (preferably the Microsoft Access database) within the pre-defined data structure in the vendor description column. The data extraction involves a temporary search for "hit" containing the price, description, and results from previous searches. The information about the product is such that the correct local character string of the language "B" typed by the user in step 312 of Fig. 15C is stored in the server 22 stored in the server 22 or the memory. Zhao Nei. Because the language "B'' character string is a special local language' any "hit" found will be recognized using the local language ''Β' and the character string with the language ''Β') Used by the seller's website. Recalling step 7 of Figure 14, on the occasion of the learning process, after the semantics of Figure 2 distinguishes that the learner agent 18 has learned the burden from the online vendor, the vendor description is stored in the offline database 24 (preferably The Vendor Description column 28 in Microsoft Access 56 1280488 V. Inventive Note (54) Database or Database Server 22 (preferably Microsoft SQL Server Compliance Server). It is inefficient to retrieve the vendor description data each time the buyer agent 20 receives a request from an online human buyer or user, if such a description is semantically different, the buyer agent 20 first requests it. The desired search-consistent_extraction will only be retrieved from the database (preferably a Microsoft Access database) or the SQL-compliant data server 22. The vendor description will then be stored in memory or cache memory to allow more simultaneous use by the same or new user in other later semantics to resolve the buyer agent 20 request. The vendor description in memory or cache memory will be automatically updated once a day. Step 3: Under the description of the vendor being used, the system of the present invention creates a multi-execution line and simultaneously deletes a number of semantically distinguished learner agents to contact various designated online vendors via the global information network. Step 4: The use of this multi-execution line methodology is preferably built on DCOM provided by Microsoft Corporation. Each semantically distinguishes the buyer's agent 2 from the product information provided by the human buyer or user and virtually presses the "type" to intelligently fill in the seller's search form. Step 5: The parent and seller then return the search result webpage with the information of the requested product or an error message. Steps 6, 7 : 1280488 V. Description of the Invention (55) The search results web pages are sent back to the semantics to distinguish the buyer agent 20 via the World Wide Web. It is worth noting that several result pages can be returned to the semantics at the same time to distinguish the buyer agent 20. The semantics of the present invention distinguishes the buyer agent 20 from the online shopper/buyer online by accelerating the vendor response time and ascending the retrieved search result web page to a separate memory location by multiple execution lines. Buying the entire process of global information online on the big internet traffic. Step 8: semantically distinguishing the buyer agent 20 analyzes the returned web page based on the corresponding vendor description. Relevant information and materials are extracted from the returned web pages and displayed in the output format as shown in Figure 15B after all search results pages have been reached or searched for more than a time. Referring to Fig. 28, the user/buyer communicates with the server 22 to execute the DLL file on the server machine 22 through an Active Server Page (ASP) (NextGen.asp) as shown (NextGen). .dll). Preferably, the development of semantically distinguishing the buyer agent 20 as an Active Component produces several benefits. First of all, the overall Zhao performance can be improved. Distinguishing the buyer agent 20 in Visual C language allows the agent to be powerful and make the powerful features of ActiveX Component available. There is no need to supply a round-trip solution in HTML and scripting to match the needs of the application. Under ActiveX Component, the agent can be executed by adding a few itineraries to the HTML file of the client, and all complex processing is thrown to the server side to be executed. Second, ActiveX Component provides reusability for other applications, instead of copying similar functionality in each application module. An ActiveX Component can be created to be accessible to all Active Server Pages. In other words, it does not require all logic to be encoded in the ASP module. So this eliminates redundancy in the application. Although the semantics of the buyer agent is created within a single application, it does not hinder the ability to integrate with other applications. In addition, this feature can help significantly reduce development time.

The third 'Connecting the ASP Component to the DLL (Dynamic Link Library) file is beneficial because these files are independently compiled and linked. There is no need to recompile and relink the ASP Component to update. Therefore, speed improvements or new features of later versions can benefit ActiveX Component using DLLs. In addition, DLLs can reduce memory and disk space requirements by sharing a single copy of common code and resources between several modules. If several elements use the same static link library, several identical copies of the library are required to be stored and executed. Then, if you run synchronously, there will be several identical copies in Memory Zhao. So it is obvious that using static link libraries will result in redundancy and wasted space. If a DLL is used instead of a static link library, only one of the code and resources needs to be copied. This is because the Internet has many simultaneous connections, so the server can be kept at a minimum load. The semantically distinguishing buyer agent 20 is preferably an ActiveX Component which is developed into a processing dll. It allows users to create one of the SRBA's objects through the World Wide Web. In order to communicate between the user and the lexical device, the ASP is used to act as a gateway between the user and the server. 59 1280488 V. Inventive Description (57). ASP is an open application environment in which HTML pages, records, and ActiveX Components are combined to create Web-based applications. In addition, it is built into the Internet Server Application Interface (ISAPI), which runs on Microsoft's Internet Information Server (IIS) product or on a layer of IIS-related Web servers. To implement ASP, Microsoft ActiveX Scripting is used, as written in Visual Basic (VB), which is used during the management of ActiveX Components. It makes the language dynamic by adding capabilities to motivate the ActiveX Component to execute as a DLL on the server. The logic of creating a semantically distinguishing one of the buyer's agents is to semantically distinguish one of the objects of the buyer's agent 20 from being created when the user begins to search for the price of the desired product. In the Active Server Page, the module in the virtual code is encoded as follows: Start sub session Set agent=sever Creat object(“NextGen·&quot;) End sub When the web page is loaded, the above module creates a semantic meaning to buy One of the agent elements 20, and NextGen is the name of the ActiveX Component. The semantics distinguishes the buyer as the name of the agent in the NextGen element. After semantically distinguishing the buyer agent 20 has been created, as shown in Figure 29. A connection is established between the user and the server. 60 1280488

V. Description of the invention (58) semantically distinguishing the buyer agent using a connectable object to maintain "one-to-one" (channel) through which the user communicates with the server, such as a user requesting to the server to compare prices, The outward interface is semantically distinguished by the buyer agent 20 as a connection (channel) through which the server communicates with the user as a response, in which the requested search result is sent back to the user. The user can access these properties and motivate the semantics to distinguish the buyer agent through IConnectionPoint. Semantic Resolution The Buyer Agent 20 uses the methodology established below: 1. OnStartPage (Unknown Agent) This methodology is used to initiate semantically distinguishing the object of the Buyer Agent, which is automatically paged when the ASP is loaded. 2. OnEndPage() This method is used to abort objects that semantically distinguish the buyer's agent, which is automatically paged when the ASP is not loaded. 3. GetSearch(BStr input, BStr ^output) This methodology is used to search for the price of the product on the Internet after the user has provided a description of the product type. Input is the user's product description, and Output is the output of the search results page. The paging syntax for this method is: OutputName=AdObjectName.GetSearch("Product Name") In the above code, AdObjectName is the instance name of the object, and "Product Name" is the buyer/user agent to compare the price of the product. The name, and OutputName are the variables that get the value returned. Refer to the following virtual code as an example, 61 1280488 V. Invention Description (59) If result = Agent Get search (“Radar detector”) The program logic login element on the server side can be used to initiate the semantics of the buyer agent. Before the object, its elements must be logged in at the server with the following command: Register path\Nextgen.dll where path is the absolute path where Nextgen.dll is saved. When semantically distinguishing the object of the buyer agent through the method of IConnectionPoint calling Get Search, the semantic machine in the server machine distinguishes the instance of the buyer agent from the dynamic link library (DLL). See figure 30. To connect to the SQL server via ODBC, you must provide the source name (DSN), identity (ID) and keywords. RETCODE is a variable that stores the value returned by the SQL server. SQL_SUCCESS indicates a successful capture. CString sDSN= “NextGen” ; CString sID= “NextGen” ; CString sPassword= “NextGen” ; RETCODE re ; rc = SQL Connect(hdbc,sDSN,“,,,sID,,,sPassword,”,,); if( Rc = = SQL_SUCCESS) else 62 1280488 V. Description of the invention (6〇) ... error handling... } Execute SQL query: In the SQL learner's desired learner profile, there must be a specific query // assign a narrative to dispose of SQLAllocStmt ( Hdbc, & hstmt): / / query statement; model is the model CString sQuery = "select''from tbl_electronic where model=+ ''model''; rc = SQLExecDirect(hstmt,sQuery,"" ); if(rc==SQL_SUCCESS) { ...take the field...

Capture Location: After the query, the vendor description will be stored in an array called vendor_ description. There are two member variables in this array: a package and a vendor URL. Re = SQLGetData(hstmt,column_number, data_type,sWrapper,sizeof(sWrapper)); if(rc= = SQL_SUCCESS) vendor_description[i].wrapper=s Wrapper, rc = SQLGetData(hstmt,column_number,data_type,sURL 63 1280488 V. DESCRIPTION OF THE INVENTION (61) sizeof(sURL)); if(rc=SQL_SUCCESS) vendor_description[i].url=sURL ; Fill in the form: If there are N vendor descriptions, the buyer agent will start N execution lines to fill in A table in each seller that is defined in the seller description. To execute each execution line, the syntax is: &quot;nNo Vendor is the number of seller descriptions Int nNoVendor=no_of_vendor_description ; For(Int nCount=0;nCount&lt;nNoVendor,nCount++) { //ThreadFillForm is the level Thread* for the implementation form ThreadFillForm ; Thread = AfxBeginThread(RUNTIME_CLASS(Thread FillForm),···) For each execution line, the time limit is preferably about 5 seconds. If the seller does not return the result within 5 seconds, the seller will be discarded this time, otherwise the result will be stored in memory for the next process. When the user enters the keyword of the purchase request for the entry of the present invention in the provided box, it is determined whether there is any relevant vendor description, that is, the vendor description p containing the keyword contains all the relevant information of the package and the URL. The seller's description is then retrieved by the offline database. Thereafter, the semantics distinguishes the buyer's agent in parallel with the searchable indicators of each online seller, and fills it in 64 1280488

5. Description of the invention (62) Write and submit to the seller's website. On the seller's website, the buyer agent will page a member function httpPost to do the job. The httpPost member function, based on the vendor description, mails the URL and form data to a vendor and sends back an HTML response as a string variable. The httpPost member function returns a Boolean value, which really indicates the success of the HTML file, and false indicates that the error occurred. If the return value is true, the generated project name and price will be extracted from the HTML file. The process of mailing the form is shown in Figure 31. In step 1002, a C internet Session object for the section is created. The Cinternet Session level is connected to a server by an internet segment. Typically this level is used earlier in the section to establish a connection to the web server. In step 1004, a CHttpConnection object is created by paging the GetHttpConnection member function of the Cinternet Session object. The CHttpConnection level establishes an HTTP connection to a server. In step 1006, a CHttpFile object's OpenRequest member function is created. The CHttpFile level allows files that are transmitted over the Internet to be processed as if they were working as a local disk archive. It works with the CHttpConnection object to read and write Internet data. Step 1008 pages the SendRequest member function of the CHttpFile object to send the POST request and the form data to the remote HTTP server. Steps 1010, 1012 and 1014 repeatedly page the Read member function of the CHttpFile object, which returns multiple pieces of response data to the program. When Read returns 0, no data is left to be retrieved. 65 1280488 V. INSTRUCTIONS (63) EXTRACTION PRICE: After obtaining the results page, semantically distinguishing the buyer agent 20 will pair each resulting web page with a generic fault template. If the web page does not match the template, the assumption is a successful search. The buyer agent 20 will then use the package for the corresponding vendor to strip the title and end from the successful web page. For example, suppose a user searches for an MD product by model MD 203 'The package is given as {7, &lt;B&gt;, &lt;/B&gt;, &lt;eight&gt;, &lt;/tarle&gt;, and the result page is The following is displayed. &lt;HTML&gt;&lt;TITLE&gt;A Simple Product Catalogs&lt;/TITLE&gt;&lt;B0DY&gt;&lt;H2&gt;MDPrice&lt;/H2&gt;&lt;TABLEB0RDER=1&gt;&lt;TRBGC0L0R=0RANGE&gt;&lt;TH&gt;ModelNumber&lt;/TH&gt;&lt;TH>PRICE(US$)&lt;/TH&gt;&lt;/TR&gt;&lt;TR&gt;&lt;TD&gt;&lt;B&gt;HM381MD&lt;/B&gt;&lt;TD&gt;&lt;;1&gt;399.95&lt;/1&gt;&lt;/TD&gt;&lt;/TR&gt;&lt;TR&gt;&lt;TD&gt;&lt;B&gt;MD2070&lt;/B&gt;&lt;TD&gt;&lt;TD&gt;&lt;1&gt;599.95&lt;lt;/l&gt;&lt;/TD&gt;&lt;/TR&gt;&lt;TR&gt;&lt;TD&gt;&lt;B&gt;MD203&lt;/B&gt;&lt;/TD&gt;&lt;TD&gt;&lt;1&gt;249.95&lt;/1&gt;&lt;/TD&gt;&lt;/TR&gt;&lt;TR&gt;&lt;TD&gt;&lt;B&gt;MDR3&lt;/B&gt;&lt;/TD&gt;&lt;TD&gt;&lt;1&gt;399.95&lt;/1&gt; 66 1280488

V. INSTRUCTIONS (64) &lt;/TD&gt;&lt;/TR&gt;&lt;/TABLE&gt;&lt;HR WIDTH=200 ALIGN=LEFT&gt;&lt;/BODY&gt;&lt;/TABLE&gt; Useful information in the package Starting at line 7 and ending at &lt;/TABLE&gt;, the semantically distinguishing buyer agent 20 will cut out useless information before extracting the model and price. The HTML file is stripped after the title and end information is: Number&lt;/TH&gt;&lt;TH&gt;PRICE(US$)&lt;/TH&gt;&lt;/TR&gt;&lt;TR&gt;&lt;TD&gt;&lt;B&gt;HM381MD&lt;/B&gt;&lt;TD&gt;&lt;1&gt;399.95&lt;/1&gt;&lt;/TD&gt;&lt;/TR&gt;&lt;TR&gt;&lt;TD&gt;&lt;B&gt;MD2070&lt;/B&gt;&lt;/TD&gt;&lt;TD&gt;&lt;1&gt;599.95&lt;1&gt;&lt;TD&gt;&lt;/TR&gt;&lt;TR&gt;&lt;TD&gt;&lt;B&gt;MD203&lt;/B&gt;&lt;/TD&gt;&lt;TD&gt;&lt;1&gt;249.95&lt;/1&gt;&lt;TD&gt;&lt;/TR&gt;&lt;TR&gt;&lt;TD&gt;&lt;B&gt;MDR3&lt;/B&gt;&lt;/TD&gt;&lt;TD&gt;&lt;1&gt;399.95&lt;/1&gt;&lt;/TD&gt;&lt;/TR&gt; Then semantically distinguishing the buyer agent 20 will use the model pairing to extract the model and price of the product. In the package, the model of the model is &lt;3&gt;&lt;^&gt;, and the price model is &lt;/1&gt;#&lt;1&gt;, where * represents the model number and # represents the price. The agent will first The model HM381MD is extracted and compared with the model "MD 203" requested by the user. If it does not match, the buyer agent 20 is semantically distinguished to find another model until it finds the model MD 203. The model is found 67 1280488

V. INSTRUCTIONS (65) Upon that, the semantics distinguishes the buyer agent 20 from using the price model to extract the first price after the model. When the model and price have been extracted, the semantically distinguishing buyer agent 20 stops extracting information from the web page and places the information in an array called array-Item[]. array_Item[] is the shared data of N execution lines, and all the execution lines can access this member variable. There is a risk that more than one execution line accesses array__Item[] at the same time, which causes an access violation. In order to protect this shared material in a consistent state, a critical segment is used to protect an execution line from modifying the data at the same time. It is declared as: CCriticalSection m_csDoor ; Before inserting an element into the array Jtem, the following line m csDoor Lock(); is added, which is used to start the key segment. All variables within the critical segment will be locked to prevent other execution lines from accessing that particular variable. After the insertion is complete, the following line m_osDoor.Unlock(); is added, which is used to indicate the end of this critical section. All locked variables will be unlocked to allow other threads to access the member variable. In doing so, the member variables of array_Item can be safely shared by all execution lines. During a specific time period, the array of stored product prices sort_item will be sorted using a quick sort method. This quick sorting method can be implemented as follows: IF left&lt;righ THEN 68 1280488

V. INSTRUCTIONS (66) BEGIN Pivot:=parition(list, left, right); Quicksort(list, left, pivot-1); Quicksort(list, pivot+1,right); END One of the structures is "key" The value is selected to be located in each reply of the code. This function then scans the entire structure in two directions. Values less than this key are passed to the left of the structure, and the larger are passed to the right. These "left to right" and "right to left" exchanges continue until a flag condition tells them to stop. An HTML file is sent back to the user, which will be stored in the member variable m_output and which displays the sorted results of the search performed by the SRBA 20. //The string used to display the content to the browser *define HTTP_HEADER "Content-type:text/htm\n\n" V shows the program to the viewer. The application is presented on the aforementioned disc. A computer program appendix that provides coded paragraphs for implementing selected features of the present invention. Specifically, in the section labeled "3.1 Learning Phase", the original program is provided as "3.1.1 Main COOSA Application Class" - a main category file for COOSA applications, " 3.1.2 Adding Seller Classes" Add a vendor category to the database, "3.1.3 COOSADoc class" is a semantically distinguishing learner agent to stimulate the display of files and screens, "3.1.4 COOSA viewing class" a learner interface and its function screen, "3.1. 5 training materials class "inspired semantics to distinguish 69 1280488 five, invention description (67) learner agent, and "3·1·6 seller class, a declaration of the rules to deal with all sellers web pages. In the label as "purchase In the stage, the original code is provided as "3 · 1 · 1 agent class, _ declares the semantics to distinguish the buyer agent, and "3.2.2 execution line processing, a word to distinguish the buyer agent processing" section. Referring now to Figures 32 through 39, the GUI or interactive agent character shopper/buyer interface used in the present invention will be described in detail. In Figure 32, the 'GUI or Interactive Agent Character Shopper/Buyer interface (1 acs/BU's "Main Menu" screen is displayed for the present invention. It will be noted that its product selection "Channel" ( The classification is provided in the upper right corner of the "Main Menu" screen. "Quick Search" is also provided on the left side of the screen. There is a box below it, where the self-typing animation guides the online human user how to use the quick search option. The left screen panel also provides a set of boxes for members to log in as a temporary trial or lifetime membership. (Most of the entry functions of the present invention are expired until the user passes the authentication.) A set of links is provided to the online seller in the lower left corner. It has been registered at the entrance of the present invention, and on the right side, a large message box is marked as "Feedback" which is provided for the user to type an opinion message into the e-mail server via e-mail, preferably Microsoft Corporation implements the Outlook Express e-mail word server. Figure 33 is a simplified diagram of the GUI or shopper/buyer interface used by the present invention, in which the company responds to The government-to-business icon is displayed, and it has been selected by the online buyer/user on the previous screen (not shown). However, it should be noted that this screen does not work because these companies, the so-called government The enterprise e-commerce service or platform provider strictly restricts members' access to the database of their web server by including the authentication security interface in the closed computer network environment. .

Figure 34 is a simplified diagram of the screen display of the GUI or shopper/buyer interface used in the present invention, wherein the user has selected the option provided after clicking the "Advanced Search" option on the screen of Figure 33. The finest of the company is provided. Note that in this screen, the title in the panel pivot under the labels of the five domains can be seen as "Advanced Agent is turned on". In addition, at the bottom of the screen, the user is provided with a dialog box that can be filled in to perform a search using the semantics of the present invention to distinguish the buyer agent function, but again, it is noted that this screen does not work because These companies, the so-called government-to-business e-business services or platform providers, severely restrict members' access to their Web server's database by including a certified security interface throughout the closed computer network environment. Figure 35 is a simplified diagram of the screen display of the GUI or shopper/buyer interface used in the present invention, in which the company is displayed in response to the "business to business" text icon, which is already in the previous screen (not shown) ) is selected by the online buyer/user. However, note that this is not the case, because the company, the so-called enterprise-to-business e-business service or platform provider, strictly limits the membership to its Web by including a certified security interface throughout the closed computer network environment. Access to the server's database. Figure 36 is a simplified diagram of the screen display of the GUI or the constructor/buyer interface used in the present invention, wherein the user has selected the option provided after clicking the "Advanced Search" option on the 35th screen. Details of the selected company are provided.

1280488 V. INSTRUCTIONS (69) Figure 37 is a simplified diagram of the GUI display of the GUI or shopper/buyer interface used in the present invention, wherein the selected item and its description are in response to the user selecting the "Domain A" flag. Hiding is displayed. Figure 38 is a simplified diagram of the screen display of the 31; 1 or shopper/buyer interface used in the present invention, which is listed in response to the user having clicked "Advanced Search" on the screen of Figure 37. Out of the sales item in the domain A. Figure 39 is a simplified diagram of the screen used by the invention (^忉 or the shopper/buyer interface), wherein the characteristics of the buyer agent are used to distinguish the buyer agent from being executed. The details of the search results are provided. The shopper/buyer responds to the user by requesting a search through the search parameter screen displayed at the bottom of Figure 38. It will be further understood that although the present invention has been Internet and globally The network is described as being accurate, and the present invention is equally applicable to recently introduced systems and next generation systems. For example, the wireless application development tool J2ME (Java for micro-editing) can be used to incorporate online knowledge-based multi-language and network. Domain-independent price comparison capability to mobile/wireless platforms, including 3 (} or web phone type, interactive and ultimate TV, pocket PC, handheld electronic notepad, web-enabled all-in-one (allMn-one) Pocket Synthesizers, wireless notebooks, etc. are used to deliver countless products and multi-language bonuses on the front page for 365 days a year, 7 days a week, 24 hours a day (24/7/3 65). Service to mobile workers and netizens. Further, the present invention can be used to deliver various products via wired and mobile/wireless platforms with multi-language value-added enterprise web services with such facilitation and functionality and such as price comparison, e-wallets Integrated, negotiating generation 1280488 V. Invention Description (70) Inter-processor communication-agent-A-to-A contract negotiation for a real-world simulation capability to multiple e-commerce segments, including consumer-to-business Consumer-to-consumer, business-to-business auctions, government-to-business transactions, etc. These A-to-A or A business activities will dynamically combine one of the keyboard, mouse and cusp devices in a global market structure. Built and launched on the Global Ensemble Marketplace Framework).

The nouns and terms that have been used herein are used as descriptive words and are non-limiting, and they are not intended to exclude the use of these nouns and terms to exclude features that are displayed and described, or features thereof. Modifications may be made in the field of the scope of the patent application of the present invention.

Component Labeling Table Component Number Translating Component Number Translator 10 Agent 110 Step 12 User/Buyer 120 Step 14 Online Seller 130 Step 16 World Wide Web/Internet 140 Step 18 Learner Agent 150 Step 20 Constructor Agent Step 210 Am 'Am Server 220 Step 24 Offline Library 230 Step 26 System Administrator 240 Step 27 Vendor Queue 250 Step 28 Vendor Description Queue 300 Step 100 Flow 囷 310 Step Name 73 1280488 V. Description of Invention (71) Component Labeling Table Component Number Translating Component Number Translated Name 312 Step 320 Step 330 Step 340 Step 350 Step 358 Step 360 Step 370 Step 380 Step 390 Step 1002 Step 1004 Step 1006 Step 1008 Step 1010 Step 步骤 Step 1012 Step 1014 Step 74

Claims (1)

Amendment 5 20 irritating members expressly - whether the original substance is changed after the amendment is amended, 1280488, _ Ί 1 ϋ ϋ 1 六 六 六 六 六 六 六 六 六 六 六 六 93 93 93 93 93 93 93 93 93 93 93 93 93 93 93 93 93 93 A method for interconnecting a computer network to select a type of instant online search processing, the method comprising the steps of: a. combining a plurality of website descriptions on a plurality of connected computer networks, each of the plurality of websites including The following are: (1) the URL used by the website; (ii) the search format URL used by the website; (iii) the generalization rules of how the selected information type on the website is organized; (IV) the information corresponding to the selection a sample of the type of website extracted; and (v) a description of the domain found on the website; b. for receiving a specific type of information from the online user; C is identified by the description of the website a specific information type network d. Use a website description for each identified website to construct a search request for the specific information type; e. Present the constructed website to the identified website Requesting a request; * f. receiving the search result from the identified website; and extracting the information corresponding to the specific information type of 4 in the local language of the website when the received search result is correctly positioned and displaying the extracted Information to the user. The method described in the first paragraph of the patent, wherein the generalized rule 75 1280488 6. The scope of the patent application includes a defining feature that accurately determines the occurrence of the type of information selected in the website. 3. The method described in claim 1 of the patent application, wherein the search for the proposed construction is ‘steps·multiple execution lines. 4. A method for synthesizing website information, which is used for real-time online search processing of information selection types interconnected with a computer network, the method comprising the following steps: a. collecting information on websites of several interconnected computer networks , including: i. URL of each of several websites; ϋ·Search form URL for each of several websites; πι·sample page containing the selected information type selected by each of several websites; iv. Location information related to the selected information type on the sample web page; and b_ derived from the collected information a generalized rule on how the selected information pattern on each of the several websites is organized. 5. The method of claim 4, wherein the step of deriving the generalization rule comprises the step of combining identification features, which can uniquely determine the selected information type in each of the plurality of websites. occur. == item one, one feature i) fixed tag character 'which is the boundary of the selected message in the sample web page; 11· further sample pages are taken by the website; and 128〇488, patent application scope Ill. Confirm the marked tokens for the further sample pages that were retrieved. 7. The method of claim 6, wherein the step of determining a tag string comprises the steps of: i arranging, for each selected information pattern, a possible string of tagged characters from the sample web pages Queue Ii·Compare the substring of the compiled character of one of the sample pages of 5H and the like, and the string of the tagged characters of the compiled character for the other sample webpage, and modify the compiled string to Aligning all of the compared sample web pages; and further wherein the step of confirming the steps comprises repeating the comparing step for a sample web page that is further retrieved by the website at the same time. 8. A method for real-time online search processing for information selection type of interconnected computer network, the method comprising the following steps: a. Accessing an offline database with one of the vendor descriptions for several seller websites on the interconnected computer network, the buyer description with information about each of the several seller websites includes: i. URL of each of several websites; Search form URL for each of several websites; ui. Description of the domain found in each of several websites; IV. General rules regarding how product information is organized on each of several websites; V· A sample of price and product information retrieved by several vendors; b. A price comparison request is received by an online user for the desired product; 77 1280488 Sixth, the scope of application for patents 10 20 c· The vendor description identifies the vendor website, which may have price information associated with the price comparison request; d· constructs a search request using the vendor description for each identified vendor website, including the corresponding Search form Url; e. Submit the constructed search request to the identified seller website; f. Extract price and product information from the search results received in response to the submitted search request, wherein the extracted price and Product information is expressed in the local language of the request; and g. displays the extracted price and product information to the user. 9. The method of claim 8, wherein the general rules include indicia that uniquely identify the occurrence of price and related information within each of the plurality of vendors. Ι〇·If the method described in claim 8 is applied, the URLs of several seller websites and the extracted price and product information are stored in a database. The method wherein the URLs of the plurality of seller websites and the extracted price and product information are stored in a database and are periodically updated. 12. The method of claim U, wherein the foxes of the plurality of vendor websites and the extracted price and product information are stored in a profile and are automatically updated daily. 13. The method of claim 8, wherein each of the online stores is different for the description. Several sellers 14 · As described in claim 13 of the patent scope, each of which is different 1280488 Patent Application Scope 10 Online store has only one vendor description stored in a database. 15. The method of claim 8, wherein the extracting step comprises the step of verifying an exact match of the search results received in response to the search request submitted with the desired product. The method of claim 8, wherein the displaying step comprises the step of displaying price and product information for the desired product of only the best selling website. 17. The method of claim 8, wherein the displaying step comprises the step of displaying price and product information for the desired product in the selectable configuration. 18. The method of claim 8, wherein the displaying step comprises the step of displaying price and product information for the desired product that has been sorted by price. 15 20 19. The method of claim 8, further comprising the step of authenticating the user comprising the secure interface. 20. The method of claim 19, wherein the security interface classifies the user as a temporary trial and a lifetime membership. 2L is the method described in claim 8, wherein the sellers of the plurality of seller websites are registered or unregistered sellers. 22. The method of claim 8, wherein the vendor descriptions are automatically constructed by an inductive learning method. 23. If applying for the method described in item 22 of section (4), the induction study can work in a multilingual environment. 24. The method of claim 22, wherein the induction learning is 79 1280488 Independent of the network. 25. The law described in Article 22 of the scope of application for patents, such as books, electronic products, and electricity. The method in which the learning method or the multiple domains of other products are summarized 26. The method described in claim 22 of the patent application scope uses a small set of training materials. Which is the induction learner 10 20 27. The method of claim 26, wherein the training material includes a product example and an URX from an online store. The method of claim 26, wherein the inductive learning method extracts and identifies data that is independent of the presentation of the online store. 29. A method for instant online search processing of information selection patterns for interconnecting computer networks, the method comprising the steps of: maintaining a separate execution database for a plurality of vendor websites connected by the interconnected computer networks Information; this information includes Fox, search form URL, domain description, and vendor description, where the vendor description includes generalization rules regarding how product information is organized on each main site; b. The information maintained in the line database is the desired product. The port is used to compare the request to process the parameter, and the price comparison is owed by the execution online user and/or semantically distinguishing the buyer agent; the one is identified by one of the several seller websites such as ° Hai Real-time price and product information 'The price and product information extracted are the local language of the website; and 80 1280488 6. The scope of application for patent d· shows the price and product information extracted to the user. The method of claim 29, wherein the step of receiving a price comparison request step comprises the step of receiving, by the user, at least one search parameter and an identity of at least one online seller on the computer network; Further, wherein the extracting step comprises the steps of: • promptly mailing a request to at least one of the plurality of online vendors using the processed parameter; and n. obtaining the search result obtained in response to the mailing step Extract information about 彳Beg and product information; iii. Sort the information obtained by price; and iv. Display the processed information on the desired products from at least a few sellers. 31. The method of claim 30, wherein the step of receiving a price comparison request is initiated by the online user. 32. The method of claim 3, wherein in the step of mailing a request, the processed parameters are search parameters, a buyer identity received by the user, and a vendor of the identified seller. A combination of the description and the URL of the identified vendor. 33. The method of claim 3, wherein the vendor description maintained in the offline database includes a step of identifying the information in the vendor website and the step of capturing the information. Use these models. 34. If the method described in the third paragraph of the patent is stipulated, the real-time price is drawn: the production=information is independent of the domain independence and has nothing to do with language independence. The method described in the third paragraph of the patent scope, wherein the processing after processing 81 1280488 6. The scope of application for patents and the steps of the materials are based on the synthetic information that will be displayed in HTML. 82 ‘