JP2005141449A - Profile generation device for information filtering, profile generation method for information filtering, and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a profile generation device or the like for information filtering which can easily generate a profile, when a profile for information filtering system is generated, and also generate a profile for adequate searching. <P>SOLUTION: The profile generating device extracts search logs based on a main key from prescribed search logs, computes the correlations among a plurality of search terms appeared in a plurality of the search logs thus extracted, selects the related terms relating to the main key based on the correlations thus computed, classifies URLs, which appear in the search logs extracted and are clicked, with a prescribed reference, computes the closeness among search terms based on the classified URLs and search terms inputted prior to the click of the URL, and generates the profiles based on the correlations to the main key and the closeness to a narrowing key. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

The present invention relates to an information filtering profile creation apparatus, an information filtering profile creation method, and a recording medium that create an information filtering profile using a search log such as a search engine.

  In a conventional information filtering system, specific documents are extracted from a document set stored in a database or a document set collected from a network or the like according to a profile in which keywords and the like are described.

  Moreover, in the conventional information filtering system, the profile is created manually. For example, in @nifty clipping service (http://www.nifty.com/iguide/clip/), a user sets a search keyword or the like and creates a profile.

  However, when a document to be searched uses a large scale and various expressions, it is extremely difficult for the user to set an appropriate keyword group. Therefore, an inappropriate document is not included in the filtered result. In this case, a necessary document may not be obtained.

  For this purpose, it is known how to perform filtering based on a set profile once, determine whether the document resulting from the filtering is appropriate or inappropriate, and modify the profile based on the determined information. (For example, refer to Patent Document 1).

  However, in the conventional profile correction method, it takes a lot of time to determine whether the result is appropriate, and if the initial profile is not appropriate to some extent, there are too many inappropriate documents in the initial filtering result, There is a high possibility that it cannot be properly corrected.

Therefore, in order to make it possible to easily create an appropriate initial profile, a related word set of keywords input by the user is automatically created based on the keywords input by the user, and the initial profile including the related words is included. In this case, a method called query expansion is known in which a search is performed based on an initial search condition input by a user, and feature words are extracted from a certain number of documents from the top in the search result ( For example, refer nonpatent literature 1).
JP 2003-157273 A Chris Buckley et al., “Using Query Zoning and Correlation Within SMART: TREC 5” 1996

  However, in the above conventional example, the high-order document in the search result is not necessarily the target document, and various topics may be written in the document. May become a word.

  That is, in the above-described conventional example, in an information filtering system that extracts specific documents from a document set stored in a database or a document set collected from a network or the like according to a profile in which keywords are described. When creating a profile for an information filtering system, there is a problem that it takes time and effort to create a profile, and there is a problem that an appropriate search cannot be performed.

The present invention relates to an information filtering system for extracting specific documents from a document set stored in a database or a document set collected from a network according to a profile in which a keyword or the like is described. When creating a profile for a system, a profile creation device for information filtering, a profile creation method for information filtering, and a recording medium that can easily create a profile and can create a profile for appropriate search are provided. It is intended to provide.

The present invention extracts a search log based on a primary key from a predetermined search log, calculates a degree of association between a plurality of search terms that appear in the extracted plurality of search logs, Based on the calculated relevance, related words related to the primary key are selected, URLs that appear in the extracted search log and clicked are classified according to a predetermined criterion, the classified URLs, Based on the search terms input prior to clicking the URL, the intimacy between the search terms is calculated, and the profile based on the relevance with the primary key and the intimacy with the refinement key Create

According to the present invention, a user selects a related word based on information in a search log and automatically creates a profile for an information filtering system simply by inputting a primary key and a narrow-down key. It is possible to reduce the time and effort required to create a profile for searching appropriately.

  The best mode for carrying out the invention is the following examples.

  A search log of an Internet search engine holds records of searches by a large number of users. In the embodiment of the present invention, whether or not the search term input by the user, the URL in the search result by the search term, the URL clicked by the user, and the time information clicked by the user are the same user Consider a search log in which a user ID or cookie that can be estimated is recorded. By using such a search log, a series of search operations of each user can be followed.

  That is, in general, a search user tends to input various keywords in order to obtain predetermined specific information. Then, keywords input at extremely close time intervals are expected to be keywords closely related to each other. A plurality of keywords used by a large number of users when clicking on the same URL are also expected to be keywords related to each other.

  The embodiment of the present invention uses the above feature in the above search behavior, and extracts the log from the search log based on the primary key input by the user. That is, based on the search history of the user who uses the word of the primary key as a search word, a log for a certain period before and after the word is input is extracted.

  Next, the degree of association between the search terms appearing in the extracted log is calculated. Based on the degree of relevance, related words included in the field specified by the primary key are specified.

  In addition, in order to narrow down to specific information in the field of the primary key, the clicked URL is classified in the extracted log, and the closeness between the input search terms is calculated before the URL is clicked. . By calculating the familiarity between the related words, it is determined whether two words used at the same time in a document related to specific information are a combination having a high possibility of being used simultaneously or a combination having a low possibility. Using the calculated familiarity, a keyword having a high familiarity with the narrow-down key input by the user is selected to generate a profile.

  FIG. 1 is a block diagram showing an information filtering profile creation apparatus 10 according to an embodiment of the present invention.

  The filtering profile creation apparatus 10 includes an input unit 1, a log extraction unit 2, a related word selection unit 3, a classification unit 4, a closeness calculation unit 5, and a profile creation unit 6. The input unit 1, the log extraction unit 2, the related word selection unit 3, the classification unit 4, the familiarity calculation unit 5, and the profile creation unit 6 include memories 1M, 2M, 3M, 4M, 5M, and 6M, respectively. doing.

  The input unit 1 includes a primary key that is a keyword for characterizing a field and a narrow-down key that is a keyword for further narrowing down information in the information field specified by the primary key in order to specify information desired by the user. input.

  The log extraction unit 2 extracts, from the search log, logs for a certain period before and after the time at which the primary key is used, from among the user logs that use the primary key as a search term.

  The related term selection unit 3 selects related terms based on the time interval in which the primary key is used, the number of people using the primary key, etc., based on the keywords included in the extracted log.

  The classification unit 4 classifies the click destination URL in the extracted log according to a predetermined criterion.

  The familiarity calculation unit 5 determines that the search terms used prior to clicking the URL in the same category are closely related to each other, and calculates the familiarity between the search terms by a predetermined method. .

  The profile creation unit 6 determines a related keyword based on the closeness key and the closeness between each related word, and generates a profile.

  FIG. 2 is a diagram illustrating an example of a search log processed by the log extraction unit 2.

  The format of the search log shown in FIG. 2 is an example, and a search time, a search word, information such as an ID and cookie for identifying a user, and a URL clicked as a result of searching using the search word are shown. Any format is acceptable as long as it contains information. In addition, it is not necessarily an actual log, and a list obtained by extracting each piece of information from the actual log may be used as a search log. Even if a search term is entered, the search result based on the entered search term is not always clicked. Therefore, if the search result is not clicked after the search term is entered, the URL field is not displayed. It is blank.

  Next, the operation of the information filtering profile creation device 10 will be described.

  FIG. 3 is a flowchart showing the operation of the information filtering profile creation apparatus 10.

  First, the input unit 1 accepts input of a primary key and a narrow-down key by a user (S1). In this case, two or more primary keys and narrowing keys may be input. The inputted primary key and narrowing key are stored in the memory 1M.

  The log extraction unit 2 examines a search term in the search log, finds a search term that matches the input primary key, and searches the user ID that entered the found search term, thereby determining the keyword of the primary key. The person who used it is specified (S2). This identified user is stored in the memory 2M.

  When there are a plurality of primary keys, for each of the primary keys, the user who uses the primary key is examined, and the log is extracted in the next step S3 by taking the union or intersection of the sets. The user who is the object is determined.

  The log extraction unit 2 extracts a log that is input within a certain period of time before and after the time when the primary key is input, from among the user logs determined in step S2, and stores the extracted log in the memory 2M. (S3). When the predetermined time is determined in advance, the predetermined time before the time when the primary key is input and the predetermined time after the time when the primary key is input need not be the same. Further, when extracting a log input within a certain time before and after the time when the primary key is input, the URL field may or may not be blank.

  The related word extraction unit 3 calculates a relevance level for the search words in the log extracted in step S3, selects words having a relevance level equal to or higher than a predetermined threshold as related words, and selects the selected related words. And stored in the memory 3M (S4).

  When calculating the degree of association, the degree of association may be calculated using the number of appearances of each search term and the number of users of each search term (the number of users specified by the user ID). That is, based on the function f (t) of the minimum time interval t between the time when each user inputs the primary key and the time when each search word is input, the obtained values are summed for all users. The following expression (1) that is a value

によって、関連度を算出するようにしてもよい。

Thus, the degree of association may be calculated.

  Note that u is the number of users, and m is the number of users using the target search term for which the relevance is being calculated. Further, the degree of association given to two predetermined words may be one or plural. The degree of association may be calculated by an expression other than the above expression (1). If there are a plurality of expressions for calculating the degree of association, the degree of association given to two predetermined words becomes plural, A threshold value is set for each of a plurality of relevance levels, and all relevance levels are equal to or higher than the threshold value. Alternatively, a search term having a relevance degree equal to or higher than a threshold may be used as a related term.

  The classification unit 4 classifies the URL in the log extracted in step S3 and stores the classified URL in the memory 4M (S5). When classifying URLs, logs whose URL field is empty are not targeted.

  When classifying URLs, URLs may be classified by grouping the same URLs, or URLs may be grouped by grouping the same domains. Further, when classifying URLs, information on URL documents is collected from a network or extracted from a search engine index or the like, a word appearance frequency vector is created, and values such as cosines are used to classify URLs. You may make it do.

  Next, the familiarity calculation unit 5 calculates the union of the related words selected in step S4 and the refinement key among the search words input prior to clicking the URL classified in step S5. Then, the familiarity between each word is calculated, and the familiarity between related words is stored in the memory 5M (S6). In other words, since the narrow-down key may not be included in the related word, the closeness between the words is calculated for the union.

  For example, the intimacy is calculated as follows.

First, for example, among the plurality of URLs described in the URL column of the search log shown in FIG. 2, the classified URL is found, and the search word in the line in which the found URL is described is found. In each set of URLs, the number of times the search word is used or the number of users is totaled for all the classified URLs. For example, the familiarity R _AB between the word A and the word B is calculated by the following equation (2) using the number of times of use or the number of users.

なお、ｉは、分類されたＵＲＬの集合を表し、ｍは、集合の数である。また、Ｎ_Ａ、Ｎ_Ｂは、それぞれステップＳ３で抽出されたログ内で、語Ａ、語Ｂを、検索語として使用した検索結果を、クリックした人数または回数（ＵＲＬ欄にＵＲＬが記載されている人数または回数）である。ｎ_Ａｉ、ｎ_Ｂｉは、ＵＲＬの集合ｉに属するＵＲＬをクリックするに先立って入力された検索語として、語Ａ、語Ｂを使用した使用人数（または、使用回数）である。

Note that i represents a set of classified URLs, and m is the number of sets. In addition, N _A and N _B are the number of people who clicked the search results using the words A and B as search words in the log extracted in step S3, respectively, or the number of clicks (URL is described in the URL column). Number or number of people). n _Ai and n _Bi are the number of people using the word A and the word B (or the number of times of use) as search terms input prior to clicking the URL belonging to the URL set i.

  However, in the case where the URL column is recorded in the log extracted in step S3 as the word of the refinement key, but no search term is described, the closeness between the refinement key and the related word is 0. Suppose that

  The profile creation unit 6 uses the familiarity calculated in step S6 to generate a profile based on the familiarity between the narrow-down key and each word, and stores the generated profile in the memory 6M (S7). When generating a profile, for example, a word of a refinement key and a list of words having a degree of association greater than a preset threshold, or a list of words of a preset number from those having a high degree of association are used as a profile. Also good.

  Further, when generating a profile, the profile may be generated using a word having a high familiarity with a word in the word list having a degree of association greater than the preset threshold value. In addition, the profile may include the numerical value of the degree of relevance with the primary key and the degree of closeness with the refinement key. Also, the numeric value of the degree of association between the word and the primary key, A predetermined function value calculated by a numerical value of the familiarity with the key may be included in the profile.

  According to the above-described embodiment, the user can generate a profile only by inputting a primary key for specifying a field and a narrow-down key for limiting to specific information of the field. The configured keyword set is appropriate.

  The above processing can be grasped as an information filtering profile creation method program.

  That is, in the above embodiment, when creating an information filtering profile for creating an information filtering profile using a predetermined search log, the above-described embodiment is based on a primary key that is a keyword for specifying an information field. From the predetermined search log, extract the search log and calculate the degree of association between the search log extraction procedure for storing in the memory and a plurality of search terms that appear in the plurality of extracted search logs, A related word selection procedure for selecting a word related to the primary key and storing it in a memory; a classification procedure for classifying URLs that appear and clicked in the extracted search log according to a predetermined criterion and store them in a memory; Based on the classified URLs and the search word information input prior to clicking the URL, the familiarity between the search words is calculated, and the memory Create a profile based on the familiarity calculation procedure to be stored, the familiarity with the refinement key, which is a keyword for further narrowing down the information in the information field specified by the primary key, and the degree of association with the primary key And an example of a program for causing a computer to execute a profile creation procedure stored in a memory.

The program may be recorded on a recording medium such as a floppy (registered trademark) disk, CD-ROM, or magneto-optical disk and executed on a computer such as a personal computer.

1 is a block diagram showing an information filtering profile creation device 10 according to an embodiment of the present invention. It is a part of an example of the search log which the log extraction part 2 processes. 4 is a flowchart showing the operation of the information filtering profile creation device 10.

Explanation of symbols

10. Information filing profile creation device,
1 ... input part,
2 ... Log extraction part,
3 ... related word selection department,
4 ... Classification part,
5 ... Intimacy calculation unit,
6: Profile creation unit.

Claims (3)

In an information filtering profile creation device for creating an information filtering profile using a predetermined search log,
An input unit for inputting a primary key that is a keyword for specifying an information field and a narrowing key that is a keyword for further narrowing down information in the information field specified by the primary key;
A search log extractor for extracting a search log from the predetermined search log based on the primary key;
The related word selection which calculates the related degree between the several search terms which appeared in the said several extracted search log, and selects the related word relevant to the said primary key based on this calculated related degree Part;
A classification unit that classifies URLs that appear and clicked in the extracted search log according to a predetermined criterion;
A closeness calculation unit for calculating closeness between search terms based on the classified URLs and search terms input prior to clicking the URL;
A profile creation unit that creates a profile based on the degree of association with the primary key and the familiarity with the narrow-down key;
An information filtering profile creation device characterized by comprising: In an information filtering profile creation method for creating an information filtering profile using a predetermined search log,
A search log extraction stage for extracting a search log from the search log based on a primary key which is a keyword for specifying an information field and storing it in a memory;
A related word selection step of calculating a degree of association between a plurality of search terms appearing in the plurality of extracted search logs, and storing the calculation result in a memory;
A classification step of classifying URLs that appear and clicked in the extracted search log according to predetermined criteria and store them in a memory;
A closeness calculating step of calculating the closeness between the search terms based on the classified URLs and the search term information input prior to clicking the URL, and storing the closeness in the memory;
Create a profile based on the familiarity with the key to narrow down the information in the information field specified by the primary key and the degree of association with the primary key, and create a profile to be stored in the memory Stages;
An information filtering profile creation method characterized by comprising: When creating a profile for information filtering that uses a predetermined search log to create a profile for information filtering,
A search log extraction procedure for extracting a search log from the search log based on a primary key which is a keyword for specifying an information field and storing it in a memory;
A related word selection procedure for calculating a degree of association between a plurality of search words appearing in the plurality of extracted search logs, selecting a word related to the primary key, and storing the word in a memory;
A classification procedure for classifying URLs that appear and clicked in the extracted search log according to predetermined criteria and store them in a memory;
A closeness calculation procedure for calculating the closeness between the search terms based on the classified URL and the search term information input prior to clicking the URL, and storing the closeness in the memory;
Create a profile based on the familiarity with the key to narrow down the information in the information field specified by the primary key and the degree of association with the primary key, and create a profile to be stored in the memory Procedures and;
The computer-readable recording medium which recorded the program which makes a computer perform.

Images