KR20100080100A - Method for searching information and computer readable medium storing thereof - Google Patents

Abstract

PURPOSE: An information searching method and a computer readable medium including a program capable of performing the same are provided to exactly select desired information by a user without exact query language input by a user. CONSTITUTION: A query is input by a searcher(S122). A topic rank between the query and a keyword or between keywords is calculated(S130). An association rank related to the query among the keywords is selected(S132). An association drawing from the query of the selected association theme is added as a weighted value so that a weighted value vector is formed. Documents related to the query are provided using the weighted value vector(S150).

Description

Method for searching information and computer readable medium storing information that includes a method for searching for information and a program capable of performing the same

The present invention relates to a computer-readable recording medium containing a method for retrieving information and a program capable of performing the same, and more particularly, to a method for forming a related subject and a related query through a subject rank, and to performing the same. The present invention relates to a computer-readable recording medium containing a program.

The use of the Internet is increasing, and accordingly, the amount of information available through the Internet is also increasing. As a result, the necessity of information retrieval becomes more and more important. However, as the amount of information increases, it becomes more difficult to find exactly the information that a searcher, a user searching for information, wants to obtain.

Initially, the results of information retrieval provided on the Internet could be obtained by human gathering and prioritizing, and providing such results according to the searcher's needs.

As the amount of information on the Internet has increased, the conventional method has reached its limit, and accordingly, a method of collecting information by a search robot and classifying and providing the information through mechanical processing has become popular. However, the result of the information retrieval by the mechanical processing does not provide exactly the results desired by the searcher, causing inconvenience that the searcher needs to find the desired information again.

SUMMARY OF THE INVENTION The present invention has been made in an effort to provide a computer-readable recording medium containing an information retrieval method capable of satisfying a searcher's needs and a program capable of performing the same.

In order to solve the above technical problem, the present invention provides a computer-readable recording medium containing the following information retrieval method and a program capable of performing the same.

The information retrieval method according to the present invention includes a data collection step of collecting a document and extracting a keyword of the collected document, a query input step of inputting a query word by a searcher, and a topic rank between the query word and the keyword or the keyword ( a related topic generating step of selecting a related topic associated with the query among the keywords by calculating a topic rank), and a related query generating step of forming a weight vector by assigning a weighted degree of association from the query word of the selected related topic. And a document retrieval step of providing documents associated with the query using the weight vector.

A subject rank TR (K, w) between the query word and the keyword and a subject rank TR (w _i , w _j ) between the keyword may be calculated by the following equation.

Where K is the query word, w, w _i , w _j is the keyword, DF (K, w) is the document frequency containing K and w together, and DF (w _i , w _j ) contains w _i and w _j together Frequency of the document, DF (w) or DF (w _j ) is the document frequency containing w or w _j , p (w) or p (w _j ) is the probability that w or w _j is in the document, α, β Is a positive real number by weight, and i and j are positive integers having different values less than or equal to the number of extracted keywords.

The selecting a related topic may include forming a connection relationship when a subject rank TR (K, w) between the query word and the keyword or a subject rank TR (w _i , w _j ) between the keyword is equal to or greater than a predetermined value. A keyword whose M or less is less than or equal to M from the query word may be selected as the related topic (M is a positive integer greater than 1).

The weight vector WM may be formed by the following equation to combine the subject rank of the association relationship formed in the query term direction of the association subject with the association subject.

Here, w _k is the k-th association subject, TR _k is the subject rank of the connection relationship formed in the query direction of the k-th association subject, N is the number of the selected association subject, k is an integer between 1 and N.

The weight vector WM combines a topic rank of a connection relationship formed in the query direction of the related subject and the related subject to reflect a depth which is the number of the connected relations between the query and the related subject. It can be formed by.

Here, w _k is the k-th association topic, TR _k is the subject rank of the connection relationship formed in the direction of the query language of the k-th association topic, depth _k is the depth that is the number of connection relationships between the query and the k-th associated topic, dTR _k Is the depth weighted topic rank of the k-th association subject, N is the number of selected association subjects, and k is an integer between 1 and N.

The data collecting step may include forming a feature keyword vector, each of which is a word vector representing a corresponding document by using the collected keywords of the document, and the document retrieving step may be performed by comparing the feature keyword vector with the weight vector. Relevant documents can be selected and provided.

The method may further include analyzing a keyword of a document selected by the searcher among the documents provided in the document searching step, and removing the keyword included in the document having a selectivity less than or equal to a predetermined threshold value from the related subject. .

According to the present invention, a related query corresponding to a specific query can be formed to provide a related subject and related document to a searcher, and the search performance can be improved by using the searcher's feedback information. In particular, by forming a related subject and an associated query associated with the query input by the searcher, the searcher can select exactly the desired information even if the searcher does not enter the correct query.

Hereinafter, an information retrieval method according to embodiments of the present invention and a computer-readable recording medium containing a program capable of performing the same will be described in detail with reference to the accompanying drawings, but the present invention is limited to the following embodiments. The present invention may be embodied in various other forms without departing from the technical spirit of the present invention. That is, specific structural to functional descriptions are merely illustrated for the purpose of describing embodiments of the present invention, and embodiments of the present invention may be embodied in various forms and should be construed as being limited to the embodiments described herein. No. It is not to be limited by the embodiments described in the text, it should be understood to include all changes, equivalents, and substitutes included in the spirit and scope of the present invention.

Terms such as first and second may be used to describe various components, but such components are not limited by the terms. The terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be referred to as the second component, and similarly, the second component may also be referred to as the first component.

When a component is said to be "connected" or "connected" to another component, it may be directly connected to or connected to that other component, but it may be understood that other components may exist in the middle. Will be. On the other hand, when a component is said to be "directly connected" or "directly connected" to another component, it will be understood that there is no other component in between. Other expressions describing the relationship between components, such as "between" and "immediately between" or "neighboring to" and "directly neighboring", will likewise be interpreted.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this application, the terms "comprise" or "include" are intended to indicate that there is a feature, number, step, action, component, or combination thereof described, and one or more other features or numbers, It will be understood that it does not exclude in advance the possibility of the presence or addition of steps, actions, components, or combinations thereof.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art. Terms such as those defined in the commonly used dictionaries are to be interpreted as having meanings consistent with the meanings in the context of the related art, and are not construed in ideal or excessively formal meanings unless expressly defined in this application. .

1 is a schematic diagram showing the configuration of an information retrieval system for implementing an information retrieval method according to an embodiment of the present invention.

Referring to FIG. 1, the information retrieval system 1 is connected through a network 100 and includes a controller 1000 and a storage 2000. The controller 1000 may include a collection unit 1100, an analysis unit 1200, an index unit 1300, a subject rank processing unit 1400, an association query generation unit 1500, a search unit 1600, and a language analysis unit 1700. , A provider 1800, a user feedback processor 1900, and the like. The storage unit 2000 may include a main storage unit 2100, an index storage unit 2200, a log storage unit 2300, and the like. The controller 1000 is connected to the Internet document 10 or the searcher device 200 through the network 100 to collect and provide information.

The collector 1100 may collect and translate the Internet documents 10 through the network 100 to generate an Internet document structure for each Internet document 10. Detailed functions and configurations of the collection unit 1100 will be described later. The Internet document 10 is a collective term for a document such as various web pages containing information including text on the Internet. Specifically, the Internet document 10 may include, for example, a typical web page, blog, news article, and the like. In addition to this, any information containing text or information that can be represented by text may be applicable. For example, various posts of a specific community (named cafes, clubs, clubs, etc.), web pages contained in various company or individual web sites, press or portal sites. News articles, etc., posts posted on various blogs, etc. may be included in the Internet document 10. In addition, the Internet document 10 may include not only text information but also various multimedia data such as pictures, videos, and music. In particular, the Internet document 10 mainly composed of multimedia data may include information consisting of text such as a title.

The analyzer 1200 may analyze the collected Internet document 10, specifically, the Internet document structures generated by the collector 1100, and generate analysis information including keywords, characteristic keyword vectors, and the like. The index unit 1300 may index the collected Internet document 10 and the analysis information to generate index information including a keyword and a characteristic keyword vector. The collected Internet document 10, the Internet document structure, analysis information, and index information may be stored in the main storage unit 2100, and the analysis unit 1200 or the index unit 1300 may be the collection unit 1100 or the analysis, respectively. The information may be received from the unit 1200 or the information stored in the main storage unit 2100 may be used. Detailed functions and configurations of the analysis unit 1200 will be described later.

The characteristic keyword vector is generated for each Internet document 100, specifically, for each Internet document structure, and represents the characteristic of the information contained in the individual Internet document 100 in the form of a word vector. The word vector contains keywords representing the characteristics of the individual Internet document 100 and the weights of the keywords. The weight is calculated using the term frequency (TF) of each keyword and the inverse document frequency (IDF), which is the inverse of the frequency of each keyword in the Internet document set. However, the frequency is the number of occurrences of a specific keyword in an individual Internet document and is a measure of how much the specific keyword represents the content of the Internet document. Reverse document frequency is also the inverse of the number of Internet documents that a particular keyword appears in an Internet document set. Keywords that appear in a few Internet documents have a greater ability to distinguish the Internet document that the keyword appears from other Internet documents. .

The topic rank processing unit 1400 may include a topic rank between keywords extracted from the Internet document 10 or a query input by the searcher through the searcher device 200 and keywords extracted from the Internet documents 10. It is possible to calculate the subject rank between and select the relevant subject from the calculated subject rank. In addition, the calculated subject rank may be stored in the index storage unit 2200. The method of calculating the subject rank will be described later.

The association query generator 1500 may generate an association query that is a weight vector using the association subject. A method of generating an association query will be described later.

The search unit 1600 may select documents associated with the query input by the searcher by comparing the related query word which is the weight vector and the feature keyword vector.

The language analyzer 1700 may perform language analysis on the query word input by the searcher through the searcher device 200 connected through the network 100. The language analyzer 1700 may determine a language of the input query word, and analyze and select a core query word when the query word is a sentence or a plurality of words.

The provider 1800 provides the searcher apparatus 200 with documents associated with the input query word selected by the searcher 1600.

The user feedback processor 1900 may store a searcher's selection result for the result provided by the provider 1800 in the log storage 2400 and reflect the result in selecting the related query.

The storage unit 2000 includes, in addition to the main storage unit 2100, an index storage unit 2200 for storing index information and a subject rank, and a log storage unit 2300 for storing searcher information including log information of searchers. do. The main storage unit 2100, the index storage unit 2200, and the log storage unit 2300 may be physically divided storage devices, or may be division units that logically divide one or a plurality of storage devices.

2 is a flowchart illustrating an information retrieval method according to an embodiment of the present invention.

1 and 2, the Internet document 10 is collected through the collection unit 1100 (S112). The collection unit 1100 may collect a specific kind of Internet document 10, or may collect a wide variety of Internet documents 10. For example, it may collect certain kinds of Internet documents 10, such as news articles or posts posted on blogs, and other types of web pages of companies or individuals, posts of specific communities, multimedia data, etc. Internet documents 10 may be collected. This may be determined according to the type of service of the information retrieval system 1 to be provided to the searcher.

The analysis unit 1200 analyzes the Internet document 10 and extracts a keyword representing each Internet document 10 (S114). Together with the keyword, a characteristic keyword vector representing the characteristic of the information contained in the Internet document 10 in the form of a word vector is formed (S116). The keyword is at least one word having meaning in the corresponding Internet document 10, and the characteristic keyword vector is a word vector including the keyword and the weight of the keyword.

On the other hand, when a searcher inputs a query word through the searcher device 200 (S122), the language analyzer 1700 analyzes the query word (S124). The searcher device 200 may search for an information retrieval system through the network 100. It can be connected with (1). The network 100 may include a wired or wireless Internet, a local LAN, an intranet, or the like. The language analyzer 1700 may classify at least one core query word by analyzing the input language, format, etc. of the query word input by the searcher, and replace the at least one core query word with a query word in a later process. Subsequently, the term "query" may be one input query word or one classified key query word, or may be a combination of at least two classified key query words. Also, the term "query" may be a query or core query in the input language, or may be translated into a main language processed by the information retrieval system 1.

Subsequently, a subject rank TR (K, w) of the extracted keyword w for the analyzed query K is calculated and a subject rank TR (w _i , w _j ) between the keywords (S130). The subject ranks TR (K, w) and TR (w _i , w _j ) can be calculated by the following equation.

K is the query word, w, w _i , w _j is the keyword, DF (K, w) is the document frequency containing K and w, and DF (w _i , w _j ) contains w _i and w _j Document frequency, DF (w) or DF (w _j ) is the document frequency containing w or w _j , p (w) or p (w _j ) is the probability that w or w _j is in the document, α, β is A positive real number by weight, i, j are positive integers having different values less than or equal to the number of extracted keywords. That is, w means any keyword among the extracted keywords, and w _i and w _j mean different keywords among the extracted keywords.

Specifically, the topic rank TR (K, w) represents the degree of association of the keyword (w) with respect to the query (K). The topic rank TR (w _i , w _j ) also indicates the degree of association of another keyword w _j with one query word w _i . DF (K, w) and DF (w) represent Document Frequency of (K and w) and (w), respectively. Similarly, DF (w _i , w _j ) and DF (w _j ) represent Document Frequency of (w _i and w _j ) and (w _j ), respectively. The document frequency refers to the number of documents in which the keywords or query terms are included in the collected documents (here, the document means individual Internet documents 10 collected). In other words, DF (K, w) is the frequency of the document containing the query (K) and the keyword (w), DF (w) is the frequency of the document containing the keyword (w). In addition, DF (wi, wj) is the frequency of documents containing two different keywords (w _i , w _j ), and DF (wj) is the frequency of documents containing keywords (w _j ).

Therefore, TR (K, w) and TR (w _i , w _j ) and DR (K, w) and DR (w _i , w _j ) have the same formulas as K and w are replaced with wi and wj. Do. Therefore, in the following explanation, the functional expressions are used as TR (x, y) and DR (x, y) to describe them (x and y are query words or keywords). Similarly, DF (w) DF (w _j ) is used as a functional expression as DF (y). In this case, p (w) or p (w _j ) will also be described as p (y).

The first part of the topic rank TR (x, y) formula, DF (x, y) / DF (y), is the ratio of documents containing x and y together in y. Therefore, the larger the value of DF (x, y) / DF (x), the higher the correlation between x and y.

However, for example, as in the case where DF (y) is 1 and DF (x, y) is 1, x and y are smaller when both the values of DF (y) and DF (x, y) are small. It can be difficult to determine if the association is really high. In other words, even though there are very few documents in which y exists along with x, it may be difficult to see that the relationship between x and y is high.

On the contrary, even if both DF (y) and DF (x, y) are close to large values, for example, the total number of documents, it may be difficult to determine whether the association between x and y is really high. In other words, it means that both x and y are frequently used, and it is difficult to give an association. For example, the case where a, the, is, of, and the like are selected as x and y in English.

The second part of the subject rank TR (x, y) formula -p (y) log (p (y)) can be used to give accuracy to the association of x and y. Hereinafter, the second part of the subject rank TR (x, y) formula, -p (y) log (p (y)), may be called an entropy part. By the entropy portion, when p (y) is 0 or 1, TR (x, y) may be 0. Accordingly, the inaccuracy may minimize the inaccuracy of the degree of association that may occur when p (y), which is a probability that y exists in documents, approaches 0 or approaches 1.

In other words, subject rank TR (x, y) improves accuracy by compensating for errors that can occur when the association between x and y is mechanically (first part of the formula) with the entropy part (second part of the formula). You can. α and β are weights given for the influence between the first and second portions of the subject rank TR (x, y) formula, respectively, and may use positive real numbers. For example, the subject rank TR (x, y) can be calculated using α = 3 and β = 2.

The subject ranks TR (K, w) and TR (w _i , w _j ) may be calculated at the stage where the query word K by the searcher is input. However, the subject rank may be calculated in advance from the keyword w stored in the index storage unit 2200 and stored together in the index storage unit 2200. Alternatively, the subject rank TR (K, w) may be calculated at the stage where the query word K is input, and the subject rank TR (w _i , w _j ) may be calculated and stored in advance.

When the subject rank TR (K, w) or TR (w _i , w _j ) is precomputed, the keyword of the collected Internet document 10 may be set to proceed every time the subject rank is extracted, and the subject rank is calculated at a specific time point. It can also be set. As such, when the subject rank between each keyword is set to be calculated at a specific time point, the time point may be determined in consideration of the amount of collected Internet documents 10 once a day, once a week, once a month, and so on. It can be set to proceed at a time when the load of the information retrieval system 1 is relatively low, that is, when the use of the searcher is low, such as at night. Thereafter, when the searcher inputs the query word K through the searcher device 200 (S122), the language analyzer 1700 analyzes the query word (S124) and calculates and stores the keyword corresponding to the query word in advance. The value can be read from the rank storage unit 2300 and used as the subject rank TR (K, w).

Therefore, the values of DF (K, w), DF (w), p (w), etc. for a specific keyword w may vary depending on a search point. Through this, search results reflecting the passage of time can be obtained.

When the subject rank TR (K, w) or TR (w _i , w _j ) is precomputed, calculations are performed every time a query is input, thereby preventing a time delay that may occur.

After calculating the subject ranks TR (K, w) and TR (w _i , w _j ), the related subjects are selected (S132). To select the related subjects, the subject ranks TR (K, w) and TR (w _i , The case where w _j ) is equal to or greater than a predetermined value is formed in a connection relationship. For example, a keyword w having a subject rank TR (K, w) equal to or larger than the predetermined value from the query word K is defined as having a connection relationship with the query word K. It is also defined that there is a connection relationship between the keyword w _i and the keyword w _j when the subject rank TR (w _i , w _j ) is equal to or greater than the predetermined value.

If such a connection relationship is defined, even a keyword that does not have a direct connection with the query word K may have a connection relationship with the query word K through another keyword. For example, even when the query word K and the keyword w ₁ have a connection relationship, and the query word K and the keyword w ₁₁ do not have a direct connection relationship, the keyword w ₁ and the keyword w ₁₁ do not have a direct connection relationship. ) Has a connection relationship, the query word (K) and the keyword (w ₁₁ ) may have a connection relationship through the keyword (w ₁ ). In this case, two connection relations may be defined between the query word K and the keyword w ₁₁ , and the number of such relations is defined as depth. In this manner, keywords having a connection relationship with the query K or less than a predetermined M, that is, a depth of M or less with the query K may be selected as the related theme. This will be explained in detail later.

An association degree is assigned to the selected association subject as a weight to generate an association query that is a weight vector (S134). An association query having a weight vector WM may be formed as follows by considering a subject rank TR (K, w) or TR (w _i , w _j ).

Here, w _k is the k-th association subject, TR _k is the subject rank of the connection relationship formed in the query direction of the k-th association subject, N is the number of the selected association subject, k is an integer between 1 and N.

Alternatively, the associative query word having the weight vector WM may be formed as in the following equation in consideration of the depth between the query word K and the selected related topic in the subject rank TR (K, w) or TR (w _i , w _j ).

Here, w _k is the k-th association topic, TR _k is the subject rank of the connection relationship formed in the direction of the query language of the k-th association topic, depth _k is the depth that is the number of connection relationships between the query and the k-th associated topic, dTR _k Is the depth weighted topic rank of the k-th association subject, N is the number of selected association subjects, and k is an integer between 1 and N.

A detailed example of forming an association query having a weight vector WM will be described in detail later.

The searcher 1600 selects the Internet documents 10 associated with the query K by comparing the weight vector WM formed for the query K with the feature keyword vector (S140). The weight vector WM and the feature keyword vector are both word vectors, and the Internet document 10 having a high correlation may be selected in consideration of the distance between the two word vectors or the angle between the two word vectors. The highly selected selected Internet document 10 is referred to as a related document.

The provider 1800 provides the searcher with the previously selected related document (S150). The associated document may be provided to the searcher a list of related documents arranged in order with the characteristic keyword vector close to the weight vector WM. As such, by forming an associated query that matches the searcher's query, the searcher can help the searcher to search for the desired information accurately without inputting the correct query.

Whether the searcher selects one of the related documents provided through the searcher device 200 may be stored in the log storage unit 2300 through the user feedback processor 1900. The user feedback by the searcher may be reflected at the time of generating the related query (S134) to reflect whether the searcher is selected (S160). More specifically, compared to other related documents, if the searcher's selection rate is low and does not exceed a predetermined threshold, it may be determined that the degree of association is low, and may be additionally reflected in the weight when generating the related query.

Alternatively, the provider 1800 may provide the searcher with the selected related theme to indicate a connection relationship with the keyword K. FIG. In this case, when the searcher selects a specific one of the related topics, the selected related topic may be set as a keyword K, and a new related topic may be selected and provided to the searcher. Alternatively, the related document may be selected and provided to the searcher as the selected related subject is inputted as the keyword K.

3 to 4 is a schematic diagram showing the configuration of the collecting unit and the analysis unit according to an embodiment of the present invention. 3 to 4, a process of storing collected Internet documents in a form that can be provided to a searcher will be described by taking blog posts as an example.

3 is a schematic view showing a configuration of a collecting unit according to an embodiment of the present invention.

Referring to FIG. 3, since the Internet document 10 and its surrounding information may be described in various languages, the language determination module 1110 may first determine a writing language such as Korean, Japanese, Chinese, and English. In addition, the collection module including the Internet document collection module 1122 and the peripheral information collection module 1124 may collect the Internet document 10 and its surrounding information together. For example, when the Internet document 10 is one post of a blog, the Internet document collection module 1122 determines an address providing a feed such as RSS / ATOM provided by the blog, thereby determining the Internet document 10. Can be collected. However, not all Internet documents 10 are provided with addresses that provide feeding. For example, since most blogs provide only a part of recent posts, in this case, the Internet document 10 may be collected through post body extraction. Similarly, the peripheral information collection module 1124 may extract and collect the peripheral information including a comment and a trackback.

The Internet document 10 and the surrounding information collected as described above are collected separately from the original form, and thus may be structured through the content restoration module 230. For example, when collecting a post of a blog, it can be analyzed and restored through the process of extracting the entire body of the post, commenting, linking trackback information, extracting existing HTML post contents, and structuring to RSS / ATOM format. In addition, according to the language determined by the language judging module 1110, the automatic translation module 1140 may translate the Internet document 10 written in a language different from the language to be provided by the automatic translation into a language to provide information about the Internet 10 and its surrounding information. Can be. The translation result of the Internet document 10 and its surrounding information restored by the content restoration module 1130 and the automatic translation module 1140 may be generated by the unit structure generation module 1150 of one Internet document 10, for example, a blog. In this case, one post can be created as an Internet document structure. The Internet document structure may be generated in a format that can be processed by a machine such as a computer, for example, XML format or RSS format.

The Internet document structure generated as described above is stored in the main storage unit 2100, and the analysis unit 1200 receives the Internet document structure from the collection unit 1100 or calls the Internet document structure stored in the main storage unit 2100 for analysis. You can work.

4 is a schematic diagram illustrating a configuration of an analysis unit according to an exemplary embodiment of the present invention.

Referring to FIG. 4, the analyzer 1200 may generate analysis information by analyzing the Internet document structure collected and generated by the collector 1100 by using a text mining technique. The analysis unit 1200 receives the Internet document structure generated by the collection unit 1100 or the Internet document structure stored in the main storage unit 2100, and performs entity name analysis in the entity name analysis module 1210 to extract the main entity names. Can be. The entity name analysis extracts words having meanings such as person name, company name, product name, service name, date, etc. by analyzing the text of the Internet document structure and can be extracted through the entity name dictionary and extraction rules. Thereafter, the extracted main entity name and information included in the Internet document structure may be statistically analyzed by the feature extraction module 1220 to extract keywords representing the Internet document structure.

The automatic classification module 1230 may automatically classify Internet document structures. Such automatic classification may be classified based on the predefined classification list 1232 and the machine learning data 1234 according to the classification list 1232. Automatic classification has high performance when the difference of classification object is clear, and classification in similar group may tend to decrease performance. Especially in the case of multi-stage classification, the analysis rate may be reduced. For example, large categories such as sports, society, and economy can be classified to some extent by the system. However, categorizing sports balls into baseball, volleyball, basketball, etc. may be less analyzed than the former. The automatic classification module 1230 may be implemented by, for example, an algorithm such as Bayesian or SVM (Support Vector Machine). In this case, the automatic classification module 1230 may automatically classify only the highest category. This is because automatic classification into multi-level categories reduces accuracy and burdens the system for machine learning.

The automatic clustering module 1240 clusters the automatically classified Internet document structures for each classification category. Auto-clustering allows the system to statistically group Internet document structures in any unit. The automatic cluster module 1240 may be implemented using, for example, a K-means algorithm. The clustered Internet document structures may be measured by the information amount index in the information amount measurement module 1250. The characteristic keyword vector generated by combining the keyword and the information quantity index can be used for searching as a word vector representing each Internet document structure. Analysis information including the extracted keyword and the generated characteristic keyword vector may be stored in the main storage unit 2100 again.

As such, the Internet document structure stored in the main storage unit 2100 may be provided in a form convenient for the searcher by the above-described information retrieval method.

5 is a conceptual diagram illustrating a process of selecting a related subject according to an embodiment of the present invention.

Referring to FIG. 5, a query word K and a plurality of keywords (w ₁ , w ₂ , w ₃ , w ₄ , w ₅ , w ₁₁ , w _12, etc., are referred to as w in the following description). Indicates an association between. Here, the association between the query word K and the keyword w or the keywords w is shown by a dotted line or a solid line. Among the correlations, solid lines indicate that TR (K, w) and TR (w _i , w _j ) form a connection relationship with the predetermined value or more, and in the case of a dotted line, TR (K, w) And when TR (w _i , w _j ) is smaller than the predetermined value, the connection relationship is not formed. That is, the query K and the keywords w ₁ , w ₂ , w ₃ , and w ₄ form a linking relationship, but the query K and the keyword w ₅ do not form a linking relationship. In addition, the keyword w ₁ and the keywords w ₁₁ and w ₁₃ form a connection relationship, but the keyword w ₁ and the keyword w ₁₂ do not form a connection relationship. Therefore, the query word K and the keyword w ₁₁ may have a connection relationship through the keyword w ₁ . In this case, the depth between the query term (K) as the keyword since the two connection relationship between the solid line (w _11), a query term (K) and a keyword (w ₁₁₎ is 2.

In contrast, the query word K and the keywords w ₃₄₁ , w ₃₄₂ , and w ₃₄₃ may have a connection relationship through the keyword w ₃ and the keyword w ₃₄ . Therefore, the depth between the query word K and the keywords w ₃₄₁ , w ₃₄₂ , and w ₃₄₃ is three. If the depth (M), which is the number of connection relationships (M) for selecting a related subject, is set to 2, the keywords (w ₁₁ , w ₁₃ , w _34, etc.) connected only by the query word (K) and the bold solid line are related subjects. However, keywords (w ₃₄₁ , w ₃₄₂ , w ₃₄₃ ) including a solid line between the query word (K) are not selected for the related subject. Through this process it is possible to select a related topic for the query language (K).

Although not shown, in the case where the connection relationship from the arbitrary keyword w to the query word K forms two or more paths, a path having a small depth between the query word K and the arbitrary keyword w may be selected. If there is a path having the same depth, one path may be selected in consideration of the value of the subject rank of each connection relationship.

In addition, when _k is the subject rank of the k-related association mentioned in the query term direction of the k-related relationship mentioned when the weight vector WM is obtained, for example, when the k-th association subject is the keyword w ₃₄ , the keyword ( w ₃ ) and the keyword (w ₃₄ ). This is because the keyword w ₃ is located in the connection relationship formed from the keyword w _{34 in the} direction of the query word K.

As described above, in the case of providing the selected related subject to the searcher, similar to the example shown in FIG. 5, the keywords w ₁₁ , w ₁₃ , and w ₃₄ connected only by bold solid lines, centering on the query word K, are illustrated. Etc.) may appear. For example, if a user selects a keyword (w ₃ ) among the provided related topics, the searcher selects a related topic again or selects a related document and provides the searcher with the keyword (w ₃ ) as a query word (K). can do. In this case, since the keywords w ₃₄₁ , w ₃₄₂ , and w ₃₄₃ have a depth between the keywords w ₃ and 2, the keywords w ₃₄₁ , w ₃₄₂ , and w ₃₄₃ may be selected as related topics and provided to the searcher.

As such, even when a query word having a comprehensive meaning is input, a related subject or related document can be efficiently provided so that a searcher can find desired information.

In addition, embodiments of the present invention can be written as a program that can be executed in a computer system. In addition, the program read from the computer-readable recording medium containing the program can be executed in the digital computer system.

Examples of computer-readable recording media include ROM, RAM, CD-ROM, DVD-ROM, magnetic tape, floppy disks, optical data storage devices, and the like, as well as carrier wave (eg, transmission over the Internet). It also includes implementations in form. The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion. And functional programs, codes and code segments for implementing the present invention can be easily inferred by programmers in the art to which the present invention belongs.

1 is a schematic diagram showing the configuration of an information retrieval system for implementing an information retrieval method according to an embodiment of the present invention.

2 is a flowchart illustrating an information retrieval method according to an embodiment of the present invention.

3 is a schematic view showing a configuration of a collecting unit according to an embodiment of the present invention.

4 is a schematic diagram illustrating a configuration of an analysis unit according to an exemplary embodiment of the present invention.

5 is a conceptual diagram illustrating a process of selecting a related subject according to an embodiment of the present invention.

<Description of main parts of drawing>

Reference Signs List 1: information retrieval system, 10: internet document, 200: searcher device, 1000: control unit, 1100: collection unit, 1200: analysis unit, 1300: index unit, 1400: subject rank processing unit, 1500: association query generation unit, 1600 : Search unit, 1700: language analyzer, 1800: provider, 1900: user feedback processor, 2000: storage, 2100: main storage, 2200: index storage, 2300: log storage

Claims (8)

A data collection step of collecting documents and extracting keywords of the collected documents; A query input step of inputting a query word by a searcher; A related topic generation step of selecting a related topic associated with the query word among the keywords by calculating a topic rank between the query word and the keyword or the keyword; An association query generation step of forming a weight vector by assigning a weighting degree of association from the query language of the selected related subject; And And a document retrieval step of providing documents associated with the query word using the weight vector. According to claim 1, The subject rank TR (K, w) between the query word and the keyword and the subject rank TR (w _i , w _j ) between the keyword are calculated by the following equation.

here, K is the query word, w, w _i , w _j is the keyword, DF (K, w) is the document frequency containing K and w, and DF (w _i , w _j ) is the document containing w _i and w _j Frequency, DF (w) or DF (w _j ) is the document frequency containing w or w _j , p (w) or p (w _j ) is the probability that w or w _j is in the document, α, β is the weight As a positive real number, i, j are positive integers having different values less than or equal to the number of extracted keywords. The method of claim 2, The selecting a related topic may include forming a connection relationship when a subject rank TR (K, w) between the query word and the keyword or a subject rank TR (w _i , w _j ) between the keyword is equal to or greater than a predetermined value. An information retrieval method (M is a positive integer greater than 1), wherein keywords having M or less keywords are selected as the related topics from the query. The method of claim 3, The weight vector WM is And a subject rank of a connection relation formed in the direction of the query term of the related subject and the related subject by the following equation.

Here, w _k is the k-th association subject, TR _k is the subject rank of the connection relationship formed in the query direction of the k-th association subject, N is the number of the selected association subject, k is an integer between 1 and N. The method of claim 3, The weight vector WM is And a subject rank of a connection relation formed in the direction of the query term of the related subject and the related subject reflecting the depth, which is the number of the connection relations between the query term and the related subject, is formed by the following equation. How to retrieve information.

Here, w _k is the k-th association topic, TR _k is the subject rank of the connection relationship formed in the direction of the query language of the k-th association topic, depth _k is the depth that is the number of connection relationships between the query and the k-th associated topic, dTR _k Is the depth weighted topic rank of the k-th association topic, N is the number of selected association topics, and k is an integer between 1 and N. According to claim 1, In the data collection step, using the keywords of the collected document to form a characteristic keyword vector, each of which is a word vector representing a corresponding document, The document retrieving step may include selecting and providing documents related to the query by comparing the characteristic keyword vector and the weight vector. The method according to claim 6, And analyzing a keyword of a document selected by the searcher among the documents provided in the document searching step, and removing the keyword included in the document having a selectivity less than or equal to a predetermined threshold value from the related subject. Characteristic information retrieval method. A computer-readable recording medium containing a program capable of performing the navigation according to any one of claims 1 to 7.

Images