KR101753768B1 - A knowledge management system of searching documents on categories by using weights - Google Patents

Abstract

A knowledge management system having a plurality of search functions according to weights, which calculates the similarity of each field of the document using a representative index word for each field and provides a search result using domain similarity for each field of the retrieved document A representative word management unit for extracting and storing representative words from the sample documents; A degree of similarity calculation unit for calculating and storing the similarity degree of each document using the representative index word for each document and the degree of similarity between the document and each document; And a search unit for searching the document according to the document search request, displaying the searched document, and displaying the degree of similarity of each searched document.
By using the knowledge management system as described above, extracting a representative index word for each field and obtaining the degree of belonging to the field by using the similarity degree with the document of each field. It is possible to analyze the extent to which documents or artifacts belong more precisely in each field, thereby providing a more accurate sectoral search.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention [0001] The present invention relates to a knowledge management system,

The present invention relates to a method and apparatus for calculating a degree of similarity in each field of a document using a representative index word for each field and providing a search result using field similarity for each field, Management system.

According to the present invention, a degree of similarity between a document and a representative index in each field is obtained, and when a user searches for a document with a minimum degree of similarity in each field, And a knowledge management system having a multiple field search function by weighting.

In general, a knowledge management system is a system that supports users to find and utilize accurate data by focusing on sharing, utilization, and collecting, accumulating, sharing, and utilizing knowledge acquisition processes. It says. The knowledge management system is based on the infrastructure of information technology such as organizational infrastructure, communication network, hardware, various software and tools such as attitude toward knowledge assets, organizational knowledge evaluation / compensation system, knowledge sharing culture have.

The knowledge management system consists of three components: knowledge base, knowledge schema, and knowledge map. If a knowledge base is compared to a database that stores raw data, then the knowledge schema can be compared to a data dictionary or database schema that contains metadata about the raw data. In the knowledge schema, contents such as the type of individual knowledge, importance, synonym, key index, security level, information of creation-inquiry-update-management department, and enterprise-wide knowledge classification system are included. As the design is important when building the house, the knowledge schema must be well established before building the knowledge management system so that the work of utilizing or maintaining the stored knowledge can be efficiently performed.

Organizations such as corporations manage knowledge systematically and introduce knowledge management system to complete knowledge management based on it. Here, knowledge management refers to "all kinds of activities that acquire and combine knowledge scattered both inside and outside the organization to achieve corporate goals and systematically share them". According to the Gartner Group, which is frequently cited for knowledge management, "Knowledge management is a management methodology for creating, collecting, structuring, accessing and using an intellectual property of a company. It also includes the expertise and experiences that are contained within. " Certainly, knowledge is not contained in a specific folder or document box. Back to top From the CEO to the janitor down to the bottom of every employee's mind is knowledge and know-how. It is knowledge management that we want to take out and systematically share. Knowledge management refers to a series of processes that create value by linking knowledge management activities to the business of a company. As a result, the reason why companies adopt the knowledge management system is to take out the knowledge and know-how contained in the head of each individual, to systematically share and manage them, and to create value unique to the company.

Particularly, in organizations such as corporations, when a specific task (or project) is decided or given, the task is often progressed and completed within a certain period of time. At the same time, it will refer to a large number of documents or materials while producing the corresponding project or project, and also produce new documents or materials. These documents referenced or produced during the project include very valuable knowledge and know-how for the organization. Therefore, it is very important to manage the above project results with knowledge base.

In particular, projects undertaken by organizations, such as corporations, are often correlated or similar. Therefore, if a person in charge of an organization such as a company can easily retrieve the results of a project conducted in the past and can accurately search results such as a desired document, it is possible to utilize these search results to obtain faster and higher quality results of the project will be.

The knowledge management system for this purpose is composed of a collection function for collecting reference documents and newly produced products, a storage function for indexing and categorizing the collected data, a sharing function for providing the accumulated knowledge easily, And provide the utilization functions to support the retrieved materials to be used in the project.

In particular, the ability to index, categorize, and store documents is crucial for retrieving documents (or output). Also, it is necessary to provide a technique to search the collected data by field, and to search documents or products accurately by field.

[Non-Patent Document 1] Ministry of Labor (2010). 2010 Knowledge Management System Improvement Project [Non-Patent Document 2] Kang, Hyojung, Kang In Tae, Yi-Ho Lee, and Yongtae Park (2002). Comparative Analysis of Type Classification and Characteristics of Commercial Knowledge Management System (KMS). 2002 Korea Industrial Engineering Association / Management Science Society Joint Conference [Non-Patent Document 3] Kim Ji Sook, Moon Hyun Jung, and Yong Tae Woo (2001). Representative Index Extraction Technique for Efficient Document Classification. Information Technology and Database Journal, 8 (1), 117-128 [Non-Patent Document 4] Hwang, Jae-young and Lee, Eung-bong (2003). A Study on Representative Index Extraction for Automatic Document Classification. Proceedings of the 10th Korea Information Management Society Conference, 55-64 [Non-Patent Document 5] Sung Hyun Woo and Park Mi Yeong (2003). A Study on the Terminology Clustering Technique of Literature Informatics Using Association Rules. Journal of the Korean Society for Library and Information Science, 37 (2), 89-105 [Non-Patent Document 6] Lee Jung Hwa, Nam Sang Yop, Moon Hyun Jung, Woo Yong Tae. Efficient Terminology Clustering Using Data Mining Techniques. 210-215 [Non-Patent Document 7] M. O. Nassar, G. Kanaan, and H. A.H. Awad (2010). Comparison between Different Global Weighting Schemes. IMECS, 2010 (1) [Non-Patent Document 8] E. Chisholm and T. G. Kolda (1999). New Term Weighting Formulas For The Vector Space Method In Information Retrieval. Computer Science and Mathematics Division [Non-Patent Document 9] Hong Sung-Jo (2013). A Study on the Completion System of Industrial Engineering Curriculum. Engineering Education Research, 16 (6), 78-86. [Non-Patent Document 10] M., Goldszmidt and M., Sahami (1998). A Probabilistic Approach to Full-Text Document Clustering. Technical Report, ITAD-433-MS-98-044, SRI International, 434-444 [Non-Patent Document 11] D., Lewis and W.A., Gale (1994). A Sequential Algorithm for Training Text Classifiers. In proceedings of the 17th Annual International ACM-SIGIR Conference on Research and Development in Information Retrieval, London, Springer-Verlag, 3-12. [Non-Patent Document 12] Cho Tae Yeon and Yoon Sung Pil (2006). A Study on the Development Strategy of New Product by Integrating Kano Model and QFD. Journal of the Korean Statistical Society, 8, 57-70. [Non-Patent Document 13] Kim, Soo-Yeon, and Sang-bok (2006). A Study on the Needs Identification of Audience Customers Based on the Kano Model. Proceedings of the Korean Society of Industrial and Engineering Engineering 16, 116-123. [Non-Patent Document 14] Patent Office Knowledge Management System User's Manual [Non-Patent Document 15] Noh Hee Seung and Yun Kyung Hee (2011). A Study on the Activation Plan of Police Knowledge Management System. Korean Public Administration Review, 8 (1), 243-264.

The object of the present invention is to solve the problems as described above, and it is an object of the present invention to provide a method and apparatus for calculating similarity of each field of a document using a representative index word for each field and providing field information on the retrieved document, And a knowledge management system having a multiple field search function.

It is also an object of the present invention to provide a knowledge management system having a plurality of categories of search functions based on weights, which calculates a degree of similarity between each field and a corresponding document using the frequency of appearance of representative index words in each field and the weight of representative index words System.

It is also an object of the present invention to provide a method and a system for searching a document related to a plurality of fields when a user searches for a document with a minimum degree of similarity in each field, And a knowledge management system having a multiple field search function based on a weight.

In order to accomplish the above object, the present invention provides a knowledge management system having a plurality of search functions according to weights, comprising: a representative word manager for extracting and storing representative words from sample documents; A degree of similarity calculation unit for calculating and storing the similarity degree of each document using the representative index word for each document and the degree of similarity between the document and each document; And a search unit for searching a document according to a document search request, displaying the searched document, and displaying the similarity of each searched document.

According to another aspect of the present invention, there is provided a knowledge management system having a plurality of search functions by weight, wherein the search unit displays radial graphs of degree of similarity in each field of the retrieved document, And the degree of similarity in each field is represented by the value of the directional axis of the corresponding field.

According to another aspect of the present invention, there is provided a knowledge management system having a plurality of search functions according to weights, wherein the search unit provides search by sector and provides a minimum degree of similarity for each field to be set. And searching and providing only those documents whose similarity degree in the field is equal to or greater than the minimum degree of similarity set in the relevant field.

The representative word management unit extracts words from the body text of the sample documents by performing morpheme analysis on the sample texts, and extracts words of each document And calculating word weights of the words by averaging the word weights for the document, and constructing a representative index word for each field from the high-weighted high-order words.

According to the present invention, in the knowledge management system having a multiple field search function based on weights, the representative word management unit calculates a word weight for each document based on a word frequency TF indicating the number of occurrences for the word t in the document d, and the inverse document frequency IDF indicating the degree of importance reduction when t is displayed in various documents.

Further, the present invention is a knowledge management system having a plurality of fields-based searching functions based on weights, wherein the representative word management unit calculates w ' _{t, d} as a word weight of a word t with respect to a document d by .

[Equation 1]

n is the number of different words appearing in document d, tf _{t, d} is the word frequency of word t for document d, and idf _t is the inverse document frequency for word t.

According to another aspect of the present invention, there is provided a knowledge management system having a plurality of search functions according to weights, wherein the representative word management unit performs association analysis with association rules that the upper words match words in the same document, The upper words are grouped into an association set by analysis and the association sets are classified into respective fields by a user's input so that words belonging to the association set classified into the corresponding field are configured as representative index words.

The present invention is characterized in that, in a knowledge management system having a multiple field search function based on a weight, the similarity degree calculation unit calculates the similarity degree between a representative index word for each field and the corresponding document according to the following [Equation 2].

[Equation 2]

Here, cos? (X, Y) is the degree of similarity between the document and the representative index word for each field, n is the number of representative index words per field, i is the index of the representative index word, Xi is the word weight for the document, Is the word weight of.

Further, the present invention provides a knowledge management system having a multi-field search function by weight, wherein the word weight Xi for the document is obtained by word frequency and inverse document frequency, And the reverse document frequency of the obtained word is used.

As described above, according to the knowledge management system having a multi-field search function by weight according to the present invention, by extracting a representative index word for each field and obtaining the degree of belonging to the field by the degree of similarity with each field document, It is possible to analyze the extent to which the document or the output belongs more accurately according to the field, thereby providing a more accurate field search.

BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 is a block diagram of a configuration example of an entire system for carrying out the present invention; Fig.
BACKGROUND OF THE INVENTION Field of the Invention [0001] The present invention relates to a knowledge management system and a knowledge management system.
3 is a flowchart illustrating a method of extracting representative words from a representative word management unit according to an exemplary embodiment of the present invention.
FIG. 4 is an exemplary diagram illustrating the result of extracting text content from a collection document according to an embodiment of the present invention; FIG.
FIG. 5 is an exemplary diagram illustrating an execution result of a morpheme analyzer according to an embodiment of the present invention; FIG.
6 is a diagram illustrating an example of a part of a terminology thesaurus according to an embodiment of the present invention;
7 is a table showing statistical values for a document and words therein according to an embodiment of the present invention;
FIG. 8 is a table showing a part of words having higher weights according to an embodiment of the present invention; FIG.
FIG. 9 is a table showing an example of discrimination of presence or absence of a document with respect to an upper unit according to an embodiment of the present invention; FIG.
10 is a table showing the number of association rules for support / reliability for the terminology "quality " according to an embodiment of the present invention.
11 is a table showing a portion of a primary association term set in accordance with an embodiment of the present invention.
12 is a table showing an example extracted by representative words according to an embodiment of the present invention.
13 is an exemplary view showing a search result by a search unit according to an embodiment of the present invention;
FIG. 14 is an exemplary view illustrating search results by field by a search unit according to an embodiment of the present invention; FIG.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the present invention will be described in detail with reference to the drawings.

In the description of the present invention, the same parts are denoted by the same reference numerals, and repetitive description thereof will be omitted.

First, the overall system configuration for carrying out the present invention will be described with reference to Fig.

As shown in FIG. 1A or FIG. 1B, the knowledge management system having a multiple field search function according to the present invention can be implemented as a server system on a network or a program system on a computer terminal.

As shown in FIG. 1A, an example of an overall system for implementing the present invention is composed of a user terminal 10 and a knowledge management server 30 and connected to each other via a network 20. It is also possible to further include a database 40 for storing necessary data.

The user terminal 10 is a conventional computing terminal such as a PC, a notebook, a netbook, a PDA, a mobile, a tablet, and a pellet which the user uses. A user requests a document search using the user terminal 10 to the knowledge management server 30 or receives the retrieved document or its results from the knowledge management server 30. [

The knowledge management server 30 is connected to the network 20 as a normal server, and stores representative index words and documents of each field. In addition, the knowledge management server 30 provides a search function for a document, searches documents according to a search request from the user terminal 10, and transmits the results.

On the other hand, the knowledge management server 30 may be implemented as a web server or a web application server that provides the respective services as web pages on the Internet. Further, the knowledge management server 30 can be constructed as an application or an application server. The knowledge management server 30 collects documents to form a knowledge base, and provides the user with a search for the documents. At this time, a function of searching and providing a document as one component of the knowledge management server can be constructed.

The database 40 is a conventional storage medium for storing data required by the knowledge management server 30, and stores data such as a representative index word and a field for document classification, or builds and stores classified documents as a knowledge base.

1B, another example of the entire system for carrying out the present invention is constituted by a knowledge management apparatus 30 of a program type installed in the computer terminal 13. [ That is, each function of the knowledge management device 30 is implemented as a computer program and installed in the computer terminal 13, receives a search request or the like through the input device of the computer terminal 13, And outputs or stores the result through the output device of the computer terminal 13. [ On the other hand, data required by the knowledge management device 30 is stored in a storage space such as a hard disk of the computer terminal 13 and used.

That is, the knowledge management system having a plurality of search functions according to the weight according to the present invention can be implemented as a program system on a computing device such as a server system or a computer terminal on a network.

Next, a knowledge management system having a multiple-field search function according to an embodiment of the present invention will be described with reference to FIG.

As shown in FIG. 2, the knowledge management system having a plurality of search functions by weight according to the present invention includes a representative word management unit 31 for extracting representative words and storing representative words, A similarity calculation unit 32, and a search unit 33 for searching a document according to a document search request and providing a result of the search. In addition, a database 40 for storing data is added.

First, the representative word management unit 31 extracts words from the sample documents and extracts representative words or representative index words from the extracted words.

As shown in FIG. 3, the representative word management unit 31 includes a body text extraction step S10, a step S20 of extracting words from the body text, a step S30 of calculating weight values of the extracted words, , And extracting representative index words for each field from the extracted words (S40), thereby extracting representative words (or representative index words).

First, the body text of the document is extracted from the sample documents (S10). That is, in the collected sample document, only the contents of the text of the document are extracted as text.

The sample documents are used for extracting representative words, and some of the documents stored in the database 40 may be used. Documents are related documents, papers, project outputs, etc. In order to extract a representative index word (or a representative word), it is possible to analyze a sufficient amount of documents belonging to each field, and to extract a representative index word for each field from the documents.

Preferably, the documents are extracted using Apache Tika. Apache Tika is an API (Application Program Interface) that provides textual text and meta information in a specific document. In other words, Apache Tika is a library that provides document type detection and content extraction from various file formats. Apache Tika supports a variety of documents, including PDF, Microsoft Office documents, and text (txt).

The text or document contents extracted from the collected documents are saved as a text file, with special characters such as *, &, ^, and spaces removed. 4 shows the execution result of the Apache Tika.

Next, words (or index words) are extracted from the body text (S20). Specifically, the morpheme is analyzed in the body text of the document, the abbreviations are removed, and the synonyms are processed by referring to the terminology dictionary. That is, the word extracting step S20 includes a morpheme analysis step S21, a stopword removal step S22, and a synonym processing step S23.

Specifically, the morpheme is analyzed from the contents of the stored text of the document, and the phrase unit and the part-of-speech are distinguished (S21). The morpheme analyzer is applied to the contents of the document body stored in a text form to divide the contents of the text into morphemes.

Preferably, commercially available tools such as a KOMORAN manufactured by SHINEWARE are used as the morpheme analyzer. In addition, any morphological analyzer such as HAM produced by Professor Kang Seung-sik (Kookmin Univ.) Or Koho morphological analyzer manufactured by IDS of Seoul National University can be applied. The morpheme analyzer is used to distinguish the word units and parts of speech. 5 shows the result of execution by the morpheme analyzer.

Next, the stopwords are removed from the separated morphemes (S22). In other words, the morphemes that are separated by the morpheme analyzer remove the insignificant words which are not meaningful as index words. Removal of abbreviations removes all other parts of speech, including verbs, verbs, conjunctions, and adjectives, except nouns and compound nouns stored in the morpheme analyzer.

For example, 'and', 'equal', and so on, which are not required as index words, are abbreviated and should be removed. Specifically, the analysis of the sentence "Production management is an important subject for improving corporate production in the field of industrial engineering" is analyzed with a stemmer analyzer. The results of the analysis are as follows: production management + industrial engineering + field + '+' Improvement '+' subject '. As in the example sentence above, except for compound nouns and nouns, all are excluded.

Next, the terminology dictionary is used to process a synonym for a terminology for a word (S23). That is, words (or terms) displayed in the same meaning but in different forms are treated as the same word or the same term. Preferably, a representative word is selected from among a plurality of words having the same meaning, and all words or terms having the same meaning as the representative index word are processed as representative words or representative words.

Synonym processing is a necessary part of the document classification process. For example, "Supply Chain Management, Supply Chain Management, SCM, Supply Chain Management" are the same jargon. Synonym processing is required to handle this in the same terminology.

Preferably, the terminology uses a terminology dictionary. That is, a terminology thesaurus is produced based on the terminology of the terminology dictionary. A thesaurus is a dictionary that shows the relationship between keywords (index words) for data retrieval, that is, a synonym, a subordinate word, a related word, and the like. Figure 6 shows a portion of a terminology thesaurus.

As shown in FIG. 6, there are words having the same meaning but different forms, such as kanban, signboard, kanban, and kanban system. It refers to industrial engineering terminology dictionary and treats words with the same meaning but different forms as one representative word. That is, integrate kanban, signboard, and kanban into kanban system and treat kanban system as four words in total. In other words, words that have the same meaning but different forms are processed.

Next, the weight of the extracted words is calculated (S30). Weights of words for each document are calculated and averaged to calculate the weight of each word.

The word weights for the document are calculated by the TF (Term Frequency) and the Inverse Document Frequency (IDF). That is, a weight value for a word w in one document D is calculated, and expressed as a statistical value indicating how important a word among the various documents is in a specific document.

Statistical values for documents and words are as follows.

TF (Term Frequency) refers to the number of occurrences of a word t in a document _d , expressed as tf _{t, d} . This is called word frequency.

Also, DF (Document Frequency) means the number of documents including the word t, which is denoted by df _t . This is called document frequency.

In addition, the IDF (Inverse Document Frequency) indicates that the importance degree is lowered when the word t appears in a plurality of documents, and is expressed by idf _t . Also, this is called reverse document frequency.

That is, the inverse document frequency IDF can be expressed by the following equation (1).

[Equation 1]

Here, N means the total number of documents.

Therefore, the word weight in the document is calculated by the following equation (2).

&Quot; (2) "

Where w _{t, d} is the weight for word t in one document d.

The statistical numerical values for the above document can be expressed in various ways as shown in FIG.

The accuracy of the TF formulas in FIG. 7 will now be discussed. Documents with a long length generally exhibit relatively high weights compared to documents with a short length because the words appear repeatedly. In addition, since it is difficult to say that the same word appears three times in a short document and three times in a long document have the same weight, an additional document length normalization is applied to make each document equal in weight and word weight.

That is, the document is normalized with respect to the document length, and the word weight is obtained in the normalized document d. The following equation (3) shows the word weight w ' _{t, d} for the normalized document d.

&Quot; (3) "

Here, n represents the number of words (different words) appearing in the document.

The above-mentioned word weights are applied to normalize and adjust the weights of words commonly appearing in all documents.

Then, for the same word, the word weights for each document are averaged, and a weight (hereinafter, a word weight) for the word is calculated.

Next, a representative index word is constructed from the extracted words (S40).

Step S40 of constructing representative words for each field includes a step of selecting upper words having a high weight in the extracted words in step S41 and a step S42 of performing association analysis using association rules in which upper words appear in the same document (S43) of grouping the words into association sets by association analysis, classifying the association sets into the respective fields by the input of the user (S44), and correcting the words in the association set by the input of the user And constituting representative words for each field (S45).

First, upper words with high weight are selected from the extracted words. That is, the upper M words or the upper M% words having the highest weight are selected from the words extracted in step S20. Hereinafter, the selected words will be referred to as the upper word because the weight is high.

For example, through the preprocessing process, we extract 1500 words with the top 5% in descending order of TF * IDF weights for extracting representative index words from 35,000 words. The table in FIG. 7 is a table showing a part of TF * IDF weighted word extraction.

Next, association analysis is performed (S42), and upper-level words are configured as a relevance set (S43)

A value indicating whether the upper words match the words in the document is obtained, and the association analysis is performed using the obtained values. That is, if the upper word A matches X with a word in one document and the other upper word B matches the word in the document, Y, the following association rule can be displayed.

R: X -> Y

Relevance analysis applies the following apriori rule.

1) If one word set (or item set) is frequent, then all subsets of this word set are also frequent item sets.

2) If a set of words is infrequent, then all sets containing this set of words are non-frequent itemsets.

For example, the set of all words is called I = {a, b, c, d}. If {b, c, d} is a frequent item set (frequent word set) then the subset {b, c}, {b, d}, {c, d}, {b} {d} is also a frequent word set, which is called a priori rule. If {a, b} is a non-frequent word set that does not exceed the minimum support criterion, {a, b, c}, {a, b, d}, {a, b, c, Can not be a frequent word set. Using this fact, word sets that do not exceed the minimum support criterion can be easily pruned. This is called a frequent word set extraction method using a priori rules [Non-Patent Document 5].

According to the above-mentioned a-priori rules, a frequent word set I is found by judging the presence / absence of words in each document. It then looks for all non-empty subsets of I for all frequent word sets I.

As shown in the table of FIG. 10, the association rules are variously output according to the changes of the minimum support and the reliability. Here, the degree of support refers to the number of documents in which the word pairs forming association rules in the entire document appear at the same time. If the degree of support is too low, the association rules are satisfied even for words that are not highly related, forming an excessive number of clusters.

Specifically, the support 10, which is about 3.3% of the document 300, is set as the minimum support. And reliability means the ratio of a and b appearing at the same time based on a word in association rule a -> b. When the reliability is increased, the number of association rules decreases according to the appearance frequency of b. Therefore, preferably, the degree of support and reliability are set to a static value of 10/55.

Specifically, to use the data mining tool IBM SPSS Modeler 14.2, the top 1500 words are extracted and compared with the words in each document. That is, when the top 1500 words are matched with words in the document, they are denoted by F, and when they are not matched. The table of FIG. 9 is an example of discrimination between upper words and presence or absence of words in the document.

Through SPSS Modeler 14.2, words that are not relevant to the discipline-specific terminology are removed through a filter node.

In addition, the number of relevance rules is set by setting different degrees of support and reliability for extracting representative index words. The table of FIG. 10 shows the number of association rules for support / reliability for the terminology 'quality'. Preferably, in the present invention, the degree of support is set to 10 and the reliability is set to 55 for extracting representative index words.

Then, we construct a first association set between each word after association analysis. After constructing a set of associations, the word set is reconstructed into field areas (system analysis, production / logistics, quality / service, ergonomics, information systems, management engineering fields, etc.) [Non-Patent Document 9].

The table of Figure 11 represents a portion of the primary association set.

Next, the association sets are classified into respective fields (S44) through the user's input, the words in the association set are corrected, and the representative word or representative index word is finally extracted (S45).

Once a number of association sets have been constructed through association analysis, it is determined which field each association belongs to. At this time, the field is determined through the input of the user.

In addition, the user receives an input from a user such as a manager for more accurate association term extraction and representative index extraction. A word which is difficult to be regarded as a representative term or a representative word in the corresponding region is removed by the input of the user and words closely related to the corresponding region which is not extracted because the degree of support is below the threshold are extracted.

For example, eliminate words that are difficult to say in the "production / logistics" domain, such as suppliers, companies, and customers, extracted from the association set. In addition, terms such as delivery date, material, order, etc. that have not been extracted due to the degree of support in the association set are below the threshold, are extracted with words closely related to the field of "production / logistics".

In addition, TF * IDF extracts a certain number of representative words for each field by additionally including words or terms representing areas of each field among words not included in the upper word due to a low weight. Preferably, a total of 180 pieces of 30 pieces are extracted for each field.

FIG. 12 shows extracted words by field.

Next, the similarity calculation unit 32 obtains the representative index word for each field and the degree of similarity between the documents, and stores the field similarity for each document.

The similarity degree calculation unit 32 calculates the similarity degree between the representative index word and the corresponding document for each field.

We use cosine coefficients to calculate the similarity between representative index words and documents. The cosine coefficient can be used to measure the degree of correspondence between features of two objects to be compared [Non-Patent Document 10]. The expression of the cosine coefficient is as follows.

&Quot; (4) "

Here, X is a word weight vector for the document of the document, and Y is a weight vector of representative index words in the corresponding field. n is the number of representative index words (or representative words) per field or area, and i is an index of a representative word.

That is, Xi is a weight of the document word, and is a weight for a word having the same meaning as the representative word of the representative word weight of Yi. On the other hand, Xi is obtained by multiplying the document frequency (df) of the corresponding word in the document by the inverse document frequency (idf).

In particular, the weight Yi of the representative word is obtained by using the word weight obtained previously. In addition, the weight Xi of the document is used by obtaining a word weight for the document of Equation (2) or (3). At this time, the document frequency tf is directly obtained from the document, and the reverse document frequency idf is used as the idf of each word obtained for the sample document.

For example, if there is a chair in the ergonomic representative word, check the number of the word 'chair' in the document (frequency) and multiply the frequency by the frequency of the inverse document (idf) of the representative word 'chair' . This will result in the weight Xi of the document word. Here, i denotes an index of a representative word indicating "chair ".

That is, the degree of similarity in Equation (4) is an index of similarity of documents in the field, and indicates how much the document belongs to the field.

Next, the retrieval unit 33 receives the retrieval request, retrieves the documents, and transmits or displays the retrieval results.

As shown in FIG. 13, the search unit 33 provides a normal search function such as a keyword search. Fig. 13 shows a screen displaying the result searched by the searching unit 33. Fig.

The search unit 33 can provide past project data and external document data most similar to the search word through calculation of the cosine similarity, rather than simply searching for the presence or absence of a word.

The search results provide meta information about the document title as well as the knowledge such as the year, source, field, and document format. In addition, the highlighted portion of the keyword is displayed so that the keyword used in the document can be grasped. This allows the user to quickly search for desired knowledge.

Specifically, documents classified into a plurality of regions or fields can be retrieved, and a document including keywords can be retrieved in real time. At this time, it is possible to search by document name, year of creation, and extension, and contents of the text can be browsed, so that the user can check before downloading the document.

In addition, the search unit 33 provides a radar chart to which the similarity calculation between the representative word and the document is applied. This provides an intuitive understanding of the area of the document.

As shown in Fig. 13 or 14, in the radial graph, the directional axes represent the respective fields, and the value in each directional axis is determined by the numerical value of the degree of similarity. The larger the degree of similarity, the greater the degree of belonging to the field. It provides an intuitive view on which field the document retrieved through the radial graph etc. belongs to.

On the other hand, the search unit 33 provides a field search, and provides a minimum similarity set for each field. That is, when the user sets the minimum similarity degree for each field, only those documents whose similarity degree in the field is equal to or higher than the minimum similarity degree set in the field are searched and provided.

The invention made by the present inventors has been described concretely with reference to the embodiments. However, it is needless to say that the present invention is not limited to the embodiments, and that various changes can be made without departing from the gist of the present invention.

10: user terminal 13: computer terminal
20: network 30: knowledge management server
40: Database

Claims (9)

1. A knowledge management system having a multiple field search function by weighting,
A representative word management unit for extracting and storing representative words from the sample documents;
A degree of similarity calculation unit for calculating and storing the similarity degree of each document using the representative index word for each document and the degree of similarity between the document and each document; And
A retrieval unit for retrieving a document according to a document retrieval request, displaying a retrieved document and providing the retrieved document,
The representative word management unit extracts words by performing morphological analysis on the body text of the sample documents, calculates a word weight for each document for each extracted word, and averages word weights for the document, A weighted value is calculated, a representative index word for each field is constructed from high weighted words,

The representative word management unit performs association analysis with the association rule that the upper words appear in the same document, groups the upper words into an association set by the association analysis, and classifies the association set into each field by a user's input The words belonging to the association set classified into the field are constructed as representative index words,
The representative word management unit processes a synonym for a word using a terminology dictionary, selects a representative word among a plurality of words having the same meaning, processes all words having the same meaning as the representative word as representative words,
Wherein the search unit displays the degree of similarity of each field of the retrieved document in a radial graph, wherein the directional axis of the radial graph represents each field, the degree of similarity for each field is represented as a value of a directional axis of the field,
The representative word management unit uses the word frequency TF indicating the number of occurrences for the word t in the document d and the inverse document frequency IDF indicating the degree of importance decreasing when the word t appears in several documents, Calculated,
Wherein the representative word management unit calculates a word weight of the word t with respect to the document d by w ' _{t, d according} to the following equation (1).
[Equation 1]

n is the number of different words appearing in document d, tf _{t, d} is the word frequency of word t for document d, and idf _t is the inverse document frequency for word t.
delete The method according to claim 1,
The search unit provides search by sector and provides a minimum degree of similarity set for each field. When the minimum degree of similarity for each field is set, the search unit searches only documents having a degree of similarity higher than the minimum degree of similarity set in the field, And a search function for a plurality of fields based on the weights.
delete delete delete delete The method according to claim 1,
Wherein the similarity degree calculation unit calculates a similarity degree between a representative index word for each field and the corresponding document according to Equation (2) below.
[Equation 2]

Here, cos? (X, Y) is the degree of similarity between the document and the representative index word for each field, n is the number of representative index words per field, i is the index of the representative index word, Xi is the word weight for the document, Is the word weight of.
9. The method of claim 8,
Wherein the word weight Xi for the document is obtained by the word frequency and the inverse document frequency, and the inverse document frequency uses the inverse document frequency of the corresponding word obtained from the sample documents. The knowledge management system comprising:

Images