TWI468956B - Method and system for personalizedly sorting searched information - Google Patents

Description

Personalized search ranking method and system

The present invention relates to a search ranking method and system, and more particularly to a personalized search ranking method and system.

With the development of information technology, information has generated the phenomenon of information explosion. The recommendation system provides a good solution. The Recommendation System uses grouping and recommended techniques to reduce the amount of information and estimate the potential interest of users.

The Recommendation System is an Information Filtering (IF) mechanism proposed to reduce the additional cost added by users in the process of searching for information. The general information filtering system is also generally referred to as a recommendation system, which not only can recommend potential information, services or products that users may need according to the user's favorite, interests, behaviors or needs (Rashid et al., 2002). If the company integrates the recommendation system into the operation structure, it can bring many potential benefits to the enterprise. For example, through the recommendation system, the merchant can analyze and judge the customer's favorite behavior by obtaining the customer's past purchase or browsing records, so that the future can be recommended. The predicted reference, in turn, stimulates customers to consume to increase sales opportunities.

However, the general recommendation system only includes the user's preference as a recommendation. Therefore, even if the user does not like certain objects, it may be listed as a favorite object a few times, and the system continues to recommend similar objects to the object.

Therefore, one aspect of the present invention is to provide a personalized search ranking method for weighting a user's favorite and dislike of various attributes as a basis for recommending the search results to the user. The personalized search sorting method includes the following steps: obtaining a plurality of preset attributes. Receive a keyword. Search for several candidate information based on keywords. Receive a client identification information from a client. According to the client identification information, one of the preset attributes is selected as the favorite attribute weight and one does not like the attribute weight. A correspondence between each candidate information and a preset attribute is obtained. One candidate information weight for each candidate information is calculated according to the correspondence relationship, the favorite attribute weight, and the unloved attribute weight. The candidate information is sorted according to the candidate information weights of the candidate information, and the sorted candidate information is transmitted to the client.

One aspect of the present invention is to provide a personalized search ranking system for weighting users' preferences and dislikes for various attributes as a basis for ranking search results and recommending them to users. The personalized search ranking system includes one storage element and a processing element electrically connected to each other. The storage component stores several preset attributes. The processing component includes a keyword processing module, a weight acquisition module, a correspondence acquisition module, a weight calculation module, and a sorting module. The keyword processing module receives a keyword and searches for several candidate information according to the keyword. The weight acquisition module receives a client identification information from a client. The weight acquisition module obtains one of the preset attribute weights and one of the favorite attribute weights according to the client identification information. The correspondence acquiring module obtains a correspondence between each candidate information and a preset attribute. The weight calculation module calculates one candidate information weight for each candidate information according to the correspondence relationship, the favorite attribute weight, and the unloved attribute weight. The sorting module sorts the candidate information according to the candidate information weights of the candidate information, and transmits the sorted candidate information to the client.

The application of the present invention has the following advantages. The user of the client can be provided with the information of the keyword entered by the client, and the information that is more liked by the client can be sorted in front and is convenient for the user to read. In addition, considering the love and dislike of the candidate information, the user of the client can be prevented from being included in the favorite information by the information that is not favorite at one time, and the information that is not favorite is continuously listed in the next search. Sort by the front.

The spirit and scope of the present invention will be apparent from the following description of the preferred embodiments of the invention. The spirit and scope of the invention are not departed.

Please refer to FIG. 1 , which illustrates a flow chart of a personalized search ranking method according to an embodiment of the invention. The personalized search ranking method weights the user's love and dislike for various attributes as the basis for recommending the search results to the user. The personalized search sorting method can be implemented as a computer program and stored in a computer readable recording medium, and the computer can perform a personalized search sorting method after reading the recording medium. Computer-readable recording media can be read-only memory, flash memory, floppy disk, hard disk, optical disk, flash drive, tape, network accessible database or familiar with the art can easily think of the same The function of the computer can read the recording media.

The personalized search ranking method 100 includes the following steps: In step 110, a plurality of preset attributes are obtained. Among them, different preset attributes can be obtained according to the applied field. In an embodiment of the present invention, when the personalized search ranking method 100 is applied to the resume file search, the preset attributes obtained may be applicants of various age ranges, various academic qualification ranges, various qualification attributes, or other types of applications. Attributes. In another embodiment of the present invention, when the personalized search ranking method 100 is applied to job search information search, the preset attributes obtained may be the type of job vacancies, the applicant's requirements, the demand qualifications, and the work location. Or other types of job vacancies. In another embodiment of the present invention, when the personalized search ranking method 100 is applied to the tutor for information search, the preset attribute obtained may be the type of teaching course required for tutoring information or other types of tutoring information attributes. . In another embodiment of the present invention, when the personalized search ranking method 100 is applied to a tutor information search, the preset attribute obtained may be a tutor teacher's skill in teaching a course type or other types of home teacher information attributes. In another embodiment of the present invention, when the personalized search ranking method 100 is applied to a tutor information search, the preset attribute obtained may be a tutor's skill in teaching type, teaching experience, or other types of home teacher information. Attributes. In another embodiment of the present invention, when the personalized search ranking method 100 is applied to the outsourcing case information search, the preset attribute obtained may be the case type of the outsourcing case or other types of outsourcing case information attributes. However, in other embodiments, different preset attributes may be obtained depending on the application field or the implementation manner.

In step 120, a keyword is received.

In step 130, a plurality of candidate information are searched for based on the keywords. The search for candidate information (step 130) may further include related words related to the received keyword into the scope of the search to further increase the number of related candidate information that can be searched. For example, when the search keyword is "Administration", you can further search for related words related to the keyword "Administration" (such as "Business Administrative Assistant", "Human Resources Assistant", etc.). In addition, depending on the application field of the personalized search ranking method 100, the candidate information may be a history document, job information, tutoring information, tutor information, outsourcing case information or other types of information.

In step 140, a client identification information, such as a client account or other type of identification information, is received from a client.

In step 150, according to the client identification information, one of the preset attribute weights and one unloved attribute weight are obtained.

In step 160, a correspondence between each candidate information and a preset attribute is obtained. Referring to FIG. 2, it is an embodiment of the correspondence between candidate information 311, 312, ..., 315 and preset attributes 321, 322, 323, 324. The candidate information 311 corresponds to the preset attributes 321, 322; the candidate information 312 corresponds to the preset attributes 321, 322, 323; the candidate information 313 corresponds to the preset attributes 323, 324; the candidate information 314 corresponds to the preset attributes 323; The candidate information 315 corresponds to the preset attribute 324.

In step 170, one candidate information weight of each candidate information is calculated according to the correspondence relationship, the favorite attribute weight, and the unloved attribute weight. The at least one corresponding attribute corresponding to each candidate information in the preset attribute may be obtained according to the correspondence relationship. The favorite attribute weights of the corresponding attributes of each candidate information are subtracted from the unloved attribute weights and summed as candidate information weights for each candidate information. For example, the candidate information 311 corresponds to the preset attributes 321, 322. Therefore, the candidate information weight of the candidate information 311 is the favorite attribute weight of the preset attribute 321 minus the unloved attribute weight, and the favorite attribute weight of the preset attribute 322 is subtracted from the unloved attribute weight. However, in other embodiments, the candidate information weights of each candidate information may be calculated by other calculation methods (step 170), and are not limited to the disclosure.

In step 180, the candidate information is sorted according to the candidate information weights of the candidate information, and the sorted candidate information is transmitted to the client. The candidate information with higher candidate information weights may be sorted before the candidate information weight is lower. Thus, the client can display the sorted candidate information on its display element. In this way, the user of the client can know the information that matches the keyword that he or she inputs, and the information that is more liked by the client can be sorted in front and is convenient for the user to read. In addition, considering the love and dislike of the candidate information, the user of the client can be prevented from being included in the favorite information by the information that is not favorite at one time, and the information that is not favorite is continuously listed in the next search. Sort by the front.

Next, the client can give back its likes and dislikes for the candidate information as a basis for correcting the weight of the favorite attribute or not. Therefore, in step 190, a selection signal can be received from the client to select a plurality of candidate information as favorite information. In step 200, the favorite attribute weights of the preset attributes are modified according to the favorite information and the corresponding relationship.

In an embodiment of step 200, the preferred attribute weight of the preset attribute can be modified by means of a binary network and an iterative operation. Therefore, according to the favorite information and the corresponding relationship, modifying the favorite attribute weight of the preset attribute (step 200) may include: generating, according to the favorite information and the corresponding relationship, one of the favorite information and the preset attribute corresponding to the favorite information. network. The weighted iteration of the favorite network is repeated several times to modify the favorite attribute weight of the preset attribute. Referring to Figures 3A-3C, an embodiment of weighted iteration of the favorite binary network is described. When the selection signal received from the client selects the candidate information 311, 314, 315 as the favorite information, the favorite binary network of FIG. 3A can be generated. The favorite attribute weights of the preset attributes 321, 322, 323, and 324 are X ₁ , X ₂ , X ₃ , and X _{4 , respectively} . Therefore, in FIG. 3B, the favorite attribute weights X ₁ and X2 of the preset attributes 321 and 322 can be iteratively added to the corresponding candidate information 311 to obtain X ₁ +X ₂ ; the favorite attribute of the preset attribute 323 can be selected. The weight X _{3 is} iteratively added to the corresponding candidate information 314; the favorite attribute weight X _{4 of the} preset attribute 324 is iteratively added to the corresponding candidate information 315. Next, in FIG. 3C, the favorite attribute weights can be iterated back to the respective preset attributes. Therefore, the weights X ₁ +X _{2 of the} candidate information 311 can be equally divided and iteratively returned to the favorite attribute weights of the preset attributes 321, 322, so that the favorite attribute weights of the preset attributes 321, 322 are (X ₁ +X ₂ )/2 . Similarly, by appreciating the iteration of the binary network, the favorite attribute weights of the preset attributes 323, 324 are respectively X ₃ , X ₄ . However, in other embodiments, the iteration of the favorite binary network may be performed more times to modify the favorite attribute weight of the preset attribute, and is not limited to the disclosure. In this way, the favorite attribute weight of each preset attribute can be modified according to the preference of the user of the client for the candidate information.

In addition, unselected ones of the candidate information may be regarded as not favorite information to further correct the unloved attribute weight of the preset attribute. Therefore, in step 210, the unloved attribute weight of the preset attribute may be modified according to the unloved information and the corresponding relationship. Then, the candidate information can be sorted by the modified favorite attribute weight and the unloved attribute weight when the keyword is received next time (step 120).

In an embodiment of step 210, the unfavourable attribute weight of the preset attribute can be modified by means of a binary network and an iterative operation. Therefore, according to the dislike information and the corresponding relationship, modifying the favorite attribute weight of the preset attribute (step 210) may include: generating the dislike information and the preset attribute corresponding to the unfamiliar information according to the dislike information and the correspondence relationship. I don't like the binary network. It is not desirable to perform multiple weight iterations on the subnet to modify the default attribute weights of the preset attributes. Referring to Figures 4A-4C, it would be an embodiment that does not like weighting iterations of the subnet. Since the selection signal received from the client selects the candidate information 311, 314, 315 as the favorite information, the other candidate information 312, 313 is regarded as not enjoying the information, and can generate the dislike network of FIG. 4A. . The unnamed attribute weights of the preset attributes 321 , 322 , 323 , and 324 are respectively Y ₁ , Y ₂ , Y ₃ , and Y ₄ , and the candidate information 312 and 313 are all corresponding to the preset attribute 323 . Therefore, in FIG. 4B, the unnamed attribute weights Y ₁ , Y _{2 of the} preset attributes 321, 322, 323, and the one-and-a-half Y ₃ /2 of the unloved attribute weights of the preset attributes 323 may be iteratively added to the corresponding ones. candidate information 312 to give _{Y 1 + Y 2 + Y 3} /2; preset attribute 323 may not like the weight of the right half of the property Y _3/2 324 and a preset attributes like the attribute weights corresponding to Y ₄ is added to the iteration Candidate information 313, and get Y ₃ /2+Y ₄ . Next, in FIG. 4C, the favorite attribute weights can be iterated back to the respective preset attributes. Therefore, the weights Y ₁ +Y ₂ +Y ₃ /2 of the candidate information 312 can be equally divided and iteratively returned to the unfavorable attribute weights of the preset attributes 321, 322, so that the uncharacteristic attribute weights of the preset attributes 321, 322 are ( Y ₁ +Y ₂ +Y ₃ /2)/3. Similarly, by not loving the iteration of the binary network, the unfavorable attribute weights of the preset attributes 323 and 324 are respectively (Y ₁ +Y ₂ +Y ₃ /2)/3+(Y ₃ /2+ Y ₄ )/2, (Y ₃ /2+Y ₄ )/2. In this way, the user of the client in the candidate information is not selected as the favorite information, and the user is regarded as not enjoying the information, and the attribute weight of each preset attribute is not modified. Therefore, the user of the client can be prevented from being included in the favorite information by the information that is not favorite at one time, and the information that is not favorite is continuously listed in the previous ranking in the next search.

Please refer to FIG. 5, which is a functional block diagram of a personalized search ranking system according to an embodiment of the invention. The personalized search ranking system weights the user's love and dislike of various attributes as the basis for recommending the search results to the user.

The personalized search ranking system 400 includes a storage component 410 and a processing component 420 that are electrically coupled to each other. The personalized search ranking system 400 can be implemented on a server or other type of computer device. In one embodiment of the invention, the personalized search ranking system 400 can establish a connection with a client 500 over a network. In another embodiment of the present invention, the personalized search ranking system 400 can be implemented on the client 500, and the client 500 can perform the personalized search ranking system 400.

The storage element 410 stores a number of preset attributes. The storage component 410 can be a flash memory, a floppy disk, a hard drive, a flash drive, a magnetic tape, a database accessible by the network, or a storage element that can be easily considered by those skilled in the art to have the same function. When the personalized search ranking system 400 is applied to the resume file search, the preset attributes stored by the storage component 410 may be various age ranges, various academic qualification ranges, various qualification attributes, or other types of applicant attributes of the applicant. In another embodiment of the present invention, when the personalized search ranking system 400 is applied to job search information search, the preset attributes stored by the storage component 410 may be the type of job vacancies, the applicant's requirements, and the qualifications of the requirements. , work location or other types of job vacancies. In another embodiment of the present invention, when the personalized search ranking system 400 is applied to the tutor for information search, the preset attribute stored by the storage component 410 may be the type of teaching required for tutoring information or other types of tutoring. Solicit information attributes. In another embodiment of the present invention, when the personalized search ranking system 400 is applied to the tutor information search, the preset attribute stored by the storage component 410 may be a tutor's skill type or other type of tutor information. Attributes. In another embodiment of the present invention, when the personalized search ranking system 400 is applied to the tutor information search, the preset attribute stored by the storage component 410 may be a tutor's skill in teaching the course type, teaching experience, or other types. Tutor teacher information attributes. In another embodiment of the present invention, when the personalized search ranking system 400 is applied to the outsourcing case information search, the preset attribute stored by the storage component 410 may be the case type of the outsourcing case or other types of outsourcing case information attributes. However, in other embodiments, the storage element 410 may store different preset attributes depending on the application or the implementation.

The processing component 420 includes a keyword processing module 421, a weight obtaining module 422, a correspondence obtaining module 423, a weight computing module 424, and a sorting module 425. The keyword processing module 421 receives a keyword from the client 500 and searches for a plurality of candidate information based on the keyword. The keyword processing module 421 may further include related words related to the received keywords into the scope of the search to further increase the number of related candidate information that can be searched. For example, when the search keyword is "administrative", the keyword processing module 421 can further search for related words related to the keyword "administration" (such as "business administrative assistant", "human resources assistant", etc. ). In addition, depending on the application field of the personalized search ranking system 400, the candidate information may be a history document, job information, tutoring information, tutor information, outsourcing case information or other types of information.

The weight acquisition module 422 receives a client identification information from a client 500. The weight acquisition module 422 obtains one of the preset attribute weights and one of the favorite attribute weights according to the client identification information. The correspondence obtaining module 423 obtains a correspondence between each candidate information and a preset attribute.

The weight calculation module 424 calculates one candidate information weight for each candidate information according to the correspondence relationship, the favorite attribute weight, and the unloved attribute weight. The weight calculation module 424 can obtain at least one corresponding attribute corresponding to each candidate information in the preset attribute according to the correspondence relationship. The weight calculation module 424 may subtract the favorite attribute weights from the favorite attribute weights of the corresponding attributes of each candidate information and add them as candidate information weights for each candidate information. However, in other embodiments, the weight calculation module 424 can calculate candidate information weights of each candidate information by other calculation methods, and is not limited to the disclosure.

The sorting module 425 sorts the candidate information according to the candidate information weights of the candidate information, and transmits the sorted candidate information to the client 500. The sorting module 425 can sort the candidate information with higher candidate information weights before the candidate information weight is lower. Thus, client 500 can display the sorted candidate information on its display elements. In this way, the user of the client 500 can know the information that matches the keyword that is input, and the information that is more liked by the user can be sorted in front and is convenient for the user to read. In addition, considering the love and dislike of the candidate information, the user of the client can be prevented from being included in the favorite information by the information that is not favorite at one time, and the information that is not favorite is continuously listed in the next search. Sort by the front.

In addition, the user of the client 500 can feed back the likes and dislikes of the candidate information through the client 500 as a basis for correcting the weight of the favorite attribute or not. Therefore, the processing component 420 can further include a selection module 426 and a favorite weight modification module 427. The selection module 426 receives a selection signal from the client. The selection signal is used to select several of the candidate information as the favorite information. The favorite weight modification module 427 modifies the favorite attribute weight of the preset attribute according to the favorite information and the corresponding relationship. The favorite weight modification module 427 can modify the favorite attribute weight of the preset attribute by means of a binary network and an iterative operation. Therefore, the favorite weight modification module 427 can generate a favorite pairing network corresponding to the favorite information among the favorite information and the preset attribute according to the favorite information and the corresponding relationship. Thus, the favorite weight modification module 427 can iterate the favorite binary network several times to modify the favorite attribute weight of the preset attribute. In this way, the favorite attribute weight of each preset attribute can be modified according to the preference of the user of the client 500 for the candidate information.

In addition, the personalized search ranking system 400 can treat the unselected ones of the candidate information as unloving information to further correct the unloved attribute weights of the preset attributes. Therefore, the processing component 420 may further include a weightless modification module 428. The dislike weight modification module 428 modifies the unloved attribute weights of the preset attributes according to the dislike information and the correspondence. The weightless modification module 428 can modify the unfamiliar attribute weight of the preset attribute by means of a binary network and an iterative operation. Therefore, the weightless modification module 428 can generate the dislike information and the one of the preset attributes corresponding to the non-favorite information, according to the dislike information and the correspondence relationship. Thus, the dislike weight modification module 428 can perform multiple weight iterations on the peer network to modify the unfavorable attribute weights of the preset attributes. In this way, the user of the client 500 in the candidate information is not selected as the favorite information, and the user is deemed to be disliked, and the attribute weight of each preset attribute is modified. Therefore, the user of the client 500 can be prevented from being included in the favorite information by one time, and the information that is not favorite is continuously listed in the previous ranking in the next search.

It will be apparent from the above-described embodiments of the present invention that the application of the present invention has the following advantages. The user of the client can be provided with the information of the keyword entered by the client, and the information that is more liked by the client can be sorted in front and is convenient for the user to read. In addition, considering the love and dislike of the candidate information, the user of the client can be prevented from being included in the favorite information by the information that is not favorite at one time, and the information that is not favorite is continuously listed in the next search. Sort by the front.

Although the present invention has been disclosed in the above embodiments, it is not intended to limit the present invention, and the present invention can be modified and modified without departing from the spirit and scope of the present invention. The scope is subject to the definition of the scope of the patent application attached.

100. . . Personalized search ranking method

110~210. . . step

311, 312, ..., 315. . . Candidate information

321, 322, 323, 324. . . Preset attribute

400. . . Personalized search ranking system

410. . . Storage element

420. . . Processing component

421. . . Keyword processing module

422. . . Weight acquisition module

423. . . Correspondence acquisition module

424. . . Weight calculation module

425. . . Sorting module

426. . . Selection module

427. . . Favorite weight modification module

428. . . Do not like weight modification module

500. . . Client

The above and other objects, features, advantages and embodiments of the present invention will become more apparent and understood.

FIG. 1 is a flow chart of a personalized search ranking method according to an embodiment of the invention.

The second figure is an embodiment of the correspondence between the candidate information 311, 312, ..., 315 and the preset attributes 321, 322, 323, 324.

The 3A-3C diagram is an embodiment of weighted iteration of the favorite binary network.

4A-4C are diagrams that do not like to perform weight iteration on the subnet.

FIG. 5 is a functional block diagram of a personalized search ranking system in accordance with an embodiment of the present invention.

100. . . Personalized search ranking method

110~210. . . step

Claims (10)

A personalized search ranking method includes: obtaining a plurality of preset attributes; receiving a keyword; searching for a plurality of candidate information according to the keyword; receiving a client identification information from a client; and according to the client identification information, Obtaining one of each of the preset attributes, a favorite attribute weight and a dislike attribute weight; obtaining a correspondence between each of the candidate information and the preset attributes; according to the correspondence, the favorite attribute weights, and The candidate information weights are calculated for each of the candidate information; and the candidate information is sorted according to the candidate information weights of the candidate information, and the sorted candidate information is transmitted to the candidate information. Client. The personalization search ranking method of claim 1, further comprising: receiving a selection signal from the client, wherein the selection signal is used to select a plurality of the candidate information as the plurality of favorite information; and according to the The favorite information and the corresponding relationship are modified, and the favorite attribute weights of the preset attributes are modified. The personalized search ranking method of claim 2, wherein modifying the favorite attribute weights of the preset attributes according to the favorite information and the corresponding relationship comprises: generating the information according to the favorite information and the corresponding relationship The favorite information and one of the preset attributes corresponding to the favorite information between the favorite information, and the plurality of weight iterations of the favorite binary network to modify the preset attributes Love attribute weights. The personalized search ranking method of claim 2, wherein the unselected ones of the candidate information are regarded as a plurality of unloved information, and the personalized search ranking method further comprises: according to the non-favorite information and the Corresponding relationship, modifying the attribute attributes of the preset attributes that do not like the attributes. The personalized search ranking method according to claim 4, wherein the unloved attribute weights of the preset attributes are modified according to the non-favorite information and the corresponding relationship: according to the unloved information and the corresponding relationship Generating the unfamiliar information and one of the preset attributes corresponding to the unfavorable information, and disabling the binary network, and modifying the unworthy part of the binary network to modify the Some of these preset attributes do not like attribute weights. The personalization search ranking method of claim 1, wherein the candidate information weight for each of the candidate information is calculated according to the correspondence, the favorite attribute weights, and the unloved attribute weights: according to the correspondence Obtaining at least one corresponding attribute of each of the preset attributes corresponding to each of the candidate information; and subtracting the favorite attribute weight of the corresponding attribute of each of the candidate information from the unloved attribute weight and summing , as the candidate information weight of each of the candidate information. The personalization search ranking method according to claim 1, wherein the candidate information is a plurality of resume documents, a plurality of job information, a plurality of tutor consultation information, a plurality of tutor information, or a plurality of outsourcing case information. A personalized search and sorting system, comprising: a storage component, storing a plurality of preset attributes; and a processing component electrically connected to the storage component, wherein the processing component comprises: a keyword processing module, receiving a keyword, And searching for a plurality of candidate information according to the keyword; a weight obtaining module receives a client identification information from a client, and obtains a favorite attribute weight of each of the preset attributes according to the client identification information and One does not like the attribute weight; a corresponding relationship acquisition module obtains a correspondence between each of the candidate information and the preset attributes; a weight calculation module, according to the correspondence, the favorite attribute weights, and the Some do not like the attribute weights, calculate one candidate information weight of each of the candidate information; and a sorting module, sort the candidate information according to the candidate information weights of the candidate information, and transmit the sorted information Candidate information to the client. The personalization search ranking system of claim 8, wherein the processing component further comprises: a selection module, receiving a selection signal from the client, wherein the selection signal is used to select a plurality of the candidate information as a plurality of favorite information; and a favorite weight modification module, according to the favorite information and the correspondence, modifying the favorite attribute weights of the preset attributes. The personalized search ranking system of claim 9, wherein the unselected ones of the candidate information are regarded as a plurality of unloved information, and the processing component further comprises: a do not like the weight modification module, according to the Do not like the information and the corresponding relationship, modify the some of the preset attributes that do not like the attribute weights.