JP2010205189A - Scoring system of search result document, score calculation device, scoring method and scoring program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To increase the score of a search result document which is actually required by a user while reducing the score of an improper document, to reduce the manpower and maintenance costs, and to reduce the resources of a storage device and a calculation device. <P>SOLUTION: A scoring system of a search result document includes a score calculation part 8A which estimates the degree of probability that satisfies a user's current search request as a numerical score for each document listed up in the search result provided by a document search part 1 as a response to the user's current search request, based on the occurrence probability 5 that the user select the document as first statistical information, and the conditional occurrence probability 6A of at least one piece of attribute information held by the user or the user's search request when selecting the document as second statistical information, generated from the user's past document selection history 3. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a scoring system, a score calculation device, a scoring method, and a scoring program for calculating a score corresponding to a high probability of conforming to a user's search request for each search result document. It is suitable for use in cooperation with a business document search engine installed in

  As this type of scoring system, for example, a priority ranking system for search result documents described in Patent Document 1 as a first related technique, or a search word ranking described in Patent Document 2 as a second related technique. A calculation system or the like is known.

  First, as shown in FIG. 14, the prioritization system as the first related technique is an accompanying function of a document search system comprising a user terminal 22 and a server computer 23 connected to each other via a network 21. It is incorporated in the server computer 23. As shown in the figure, this priority order assigning system is roughly composed of a data input / output device 24, a data processing device 25, and a storage device 26.

  The data processing device 25 includes a search execution unit 27, a score calculation unit 28, a data generation unit 29, and a favorite list registration unit 30, as shown in FIG. The search execution unit 27 performs a document search according to a search request from a user (user terminal 22), generates a search result document list, and stores the score calculation unit 28 and the storage device 26 (search source data storage). Part 31). The score calculation unit 28 has a probability (possibility) of matching each user's search request for each document (hereinafter also referred to as a search result document) listed in the search result document list provided from the search execution unit 27. Perform scoring (score calculation) according to the height. The data generation unit 29 generates document information to be ranked with the help of user input. The favorite list registration unit 30 registers document information to be ranked in the favorite list with the help of user input. Further, the search execution unit 27 assigns priorities to each search result document according to the level of the score obtained by the score calculation unit 28, and responds to the search request from the user.

  Further, the storage device 26 stores a search source data storage unit 31 that stores document information to be searched, a search result document, and the like, and a score (score) of each search result document calculated by the score calculation unit 28. And a calculation result storage unit 32 to be held. Various data stored in the storage device 26 are written by the search execution unit 27 and the score calculation unit 28, and read out as necessary.

  In the prioritization system configured as described above, when the data input / output device 24 receives a search request from the first user, the search execution unit 27 searches the search source data storage unit 31 for a document. The search result document list is obtained and output to the score calculation unit 28. The score calculation unit 28 performs score calculation based on document information to be ordered and created and stored by the first user and the second user. In this score calculation, the relevance of the search result document between the first user who requests the search and the second user who creates and stores the document information to be ordered. Presence or absence is determined.

In this relevance determination, it is determined whether or not the attribute information (for example, community) held by the first and second users is the same, and if the result of the determination is the same, “relevant” is determined. If it is determined that they do not match, it is determined that there is no relevance. That is, in the system operating in the company, for example, it is determined whether or not the “affiliation department” is the same, and if it is the same, it is determined as “relevant”. When the score calculation unit 28 is “relevant”, the score calculation unit 28 viewed the search result document from the second user based on the document information to be ordered and created and registered by the second user. Assign a score and calculate priority. At this time, when the first user also creates and registers the document information to be ordered, the score calculation unit 28 sets the score viewed from the second user for the search result document. The priority is calculated by assigning a value added to the score seen by one user as a score. When the search execution unit 27 receives the score for the search result document, the search execution unit 27 outputs the search result with the priority order to the first user via the data input / output device 24.
Thus, according to the configuration of the first related technology, it is possible to prioritize search result documents according to community values.

  Next, the search word ranking calculation system as a second related technique will be described. The ranking calculation system calculates the ranking of search terms that appear in a certain period, and includes first to third means for determining first to third values that increase (or decrease) a ranking evaluation value; And a fourth means for determining an overall ranking of search terms based on the first to third values. The first means uses the search log information read from the storage means to increase a ranking evaluation value for a search word having a higher use frequency among search words appearing during the ranking evaluation period. Generate the value of. A 2nd means produces | generates the 2nd value which lowers a ranking evaluation value, so that the search word (for example, the search word in which the frequency | count of the recent input is monotonously decreasing) used many retroactively in the past N period. The third means generates a third value that increases the ranking evaluation value for a search term that has been recently used frequently (for example, a search term whose number of recent inputs has monotonously increased).

  According to the configuration of the second related technique, the value of the search table that newly appears is higher (ranked higher) than the search term that continues to appear for a long time, and thus the input frequency tends to increase recently. It is possible to increase the ranking of search words, that is, search words that can be presumed to become a topic in the future. Therefore, trend information reflecting the times can be provided to the user.

JP 2008-225792 A JP 2005-309760 A

However, in the first related technology, since the system configuration assigns priorities by relying on document information created and registered by hand, the target ratio of priority assignment is the creation registrant. There is a problem of being easily affected by the experience and qualities of
If the inexperienced person creates document information for assigning priorities, there is a possibility that document information with a strong personal preference (that is, document information with a light community color) is mixed and an erroneous document is ranked higher. So there is a problem. In order to avoid this, it is necessary to secure an expert or expert of document information creation, which is inconvenient because an increase in maintenance cost cannot be avoided.
In addition, manual selection and extraction processing of documents to be ranked and creation / registration processing of the document information are indispensable, and every time the target document is changed due to the transition of business contents, update processing is also forced. Therefore, there are also disadvantages that the process is complicated and requires a lot of work time.

  In addition, in order for the first related technology to be operated as an in-house system, at least the number of attribute values (affiliated departments) has a strong relationship with each attribute, and the document information to be ranked. Generation registration processing is required. However, an increase in registered document information corresponding to the number of attribute values is not preferable in terms of utilization efficiency of storage device resources. In particular, as the attribute values such as departmental staff are differentiated and specialized, the number of registered document information and the number of managers in charge increases, which hinders efficient use of storage devices and computing device resources, and in turn There is also a problem that the resources of the apparatus and the computing apparatus are increased.

  On the other hand, in the second related technology, since attributes such as departments in a company are not judged, there is a possibility that even a meaningful search word in a specific department may be judged as a non-conforming search word (noise) in other departments. There is. In addition, because the division of roles has been transferred from one department to another, search terms tend to increase recently in other departments, but tend to decrease recently in other departments. May occur in many cases. In such a case, it is inconvenient because a search term that is no longer important (a search term that tends to decrease recently input frequency) may remain in a higher rank from a certain department. On the other hand, from the viewpoint of other departments, there is a risk that search terms that will increase in importance (search terms that tend to increase recently) may not be ranked higher as expected. is there. Furthermore, once a trouble that increases the input frequency of non-conforming search words (noise) occurs, the system erroneously determines that this non-conforming search word is a search word that will become a topic in the future. There is also a problem that it is ranked higher and higher.

The present invention has been made in view of the above circumstances, and is an easy-to-use search that can increase the score of a search result document that a user really needs and can decrease the score of a nonconforming document (noise). It is a first object of the present invention to provide a result document scoring system, a score calculation device, a scoring method, and a scoring program.
The present invention also provides a search result document scoring system, a score calculation device, a scoring method, which can achieve a reduction in labor and maintenance costs, and can reduce storage devices and computing device resources, and The second object is to provide a scoring program.

In order to solve the above-described problem, a first configuration of the present invention relates to a search result document scoring system, and for each search result document provided from a document search means as a response to a user's current search request. , A score calculation means for estimating the degree of probability that matches the user's current search request and digitizing the score is provided.
The score calculation means calculates a score for each search result document based on the first statistical information and the second statistical information respectively generated from the user's past document selection history. The first statistical information is information regarding the probability of occurrence of the document selection by the user, and the second statistical information is at least one attribute of the current user or the search request when the document is selected. It is information regarding the conditional occurrence probability of information.
Here, the conditional occurrence probability of the attribute information, which is the second statistical information, is, for example, at least one for each search target document and the user or the search request for selecting the document. Is defined as the probability that the attribute takes each attribute value.

The second configuration of the present invention relates to a score calculation device for a search result document. As a response to the user's current search request, the user's current search request is provided for each search result document provided from the document search means. Estimate the degree of probability of conforming to, and digitize as a score.
The score calculation device calculates a score for each search result document based on the first statistical information and the second statistical information respectively generated from the user's past document selection history. The first statistical information is information regarding the probability of occurrence of the document selection by the user, and the second statistical information is at least one attribute of the current user or the search request when the document is selected. It is information regarding the conditional occurrence probability of information.

A third configuration of the present invention relates to a search result document scoring method, wherein the user's current search request is provided for each search result document provided by the document search means as a response to the user's current search request. It has a score calculation step for estimating the probability of conformity to the search request and digitizing it as a score.
In this score calculation step, a score is calculated for each search result document based on the first statistical information and the second statistical information generated from the user's past document selection history. The first statistical information is information regarding the probability of occurrence of the document selection by the user, and the second statistical information is at least one attribute of the current user or the search request when the document is selected. It is information regarding the conditional occurrence probability of information.

According to a fourth aspect of the present invention, there is provided a scoring program for a search result document, wherein the user is provided with respect to each search result document provided from the document search means as a response to the user's current search request. The degree of probability that matches the current search request is estimated, and a score calculation process is performed to quantify the score.
In this score calculation process, for each search result document, a score is calculated based on the first statistical information and the second statistical information respectively generated from the user's past document selection history. The first statistical information is information regarding the probability of occurrence of the document selection by the user, and the second statistical information is at least one attribute of the current user or the search request when the document is selected. It is information regarding the conditional occurrence probability of information.

  According to the configuration of the present invention, for each search result document, based on two or more pieces of probability information generated from the user's past document selection history, the degree of probability that matches the user's search request is estimated. Since it is digitized as a score, it is possible to improve the hit rate of prioritizing search result documents. Here, the probability information of two or more is the occurrence probability of the document selection by the user, which is considered to be highly related to each other, and the conditional occurrence probability of at least one query attribute information. If these probabilities are integrated and integrated, the accuracy of the probabilities can be significantly increased. Therefore, the score of the search result document that the user really needs can be increased while the score of the nonconforming document (noise) can be decreased. Here, the query attribute information is the conditional occurrence probability (for example, the probability that each attribute value takes for each attribute) of each attribute information that the user or the search request has when selecting a search result document. .

  According to the present invention, a document selection history storage unit for storing a user's past document selection history based on the search result provided from the document search means, and reading out the document selection history from the document selection history storage unit, two or more Probability information generating means for generating the probability information. In this way, it is possible to eliminate manual distributed registration work, so that it is possible to achieve a reduction in labor and maintenance costs, and also to reduce storage devices and computing device resources. A good system environment can be realized.

1 is a block diagram showing a configuration of a search result document scoring system according to a first embodiment of the present invention. FIG. It is a conceptual diagram which shows typically an example of the document selection log | history information table set to document selection log | history DB which comprises the scoring system. It is a conceptual diagram which shows typically an example of the occurrence probability table of the document selection set to document occurrence probability DB which comprises the same scoring system. It is a conceptual diagram which shows typically the example of the conditional occurrence probability table of the various attributes set in the attribute conditional occurrence probability DB which comprises the scoring system. It is a flowchart which shows the operation | movement procedure of the score calculation process which the score calculation part which comprises the scoring system performs. It is a flowchart which shows the operation | movement procedure of the probability DB update process which the probability DB production | generation part which comprises the scoring system performs with respect to various probability DB. It is a graph curve figure for demonstrating the weight calculation process performed in the weight calculation part of the old occurrence probability which comprises the same probability DB production | generation part. It is a block diagram which shows the structure of the scoring system of the search result document which is 2nd Embodiment of this invention. It is a conceptual diagram which shows typically an example of the generation probability table of various attributes set in attribute generation probability DB which comprises the scoring system. It is a conceptual diagram which shows typically an example of the conditional occurrence probability table of the various attributes set in attribute conditional occurrence probability DB which comprises the scoring system. It is a flowchart which shows the operation | movement procedure of the score calculation process which the score calculation part which comprises the scoring system performs. It is a flowchart which shows the operation | movement procedure of the probability DB update process which the probability DB production | generation part which comprises the scoring system performs with respect to various probability DB. It is a block diagram which shows the structure of the scoring system of the search result document which is a modification of 2nd Embodiment. 1 is a block diagram schematically showing an electrical configuration of a priority assignment system as a first related technique. FIG.

An outline of the embodiment of the present invention is as follows. In order to achieve the above object, a document selection history DB (database) 4 for storing a user's past document selection history based on a search result provided from the document search unit 1 is provided. It is preferable to provide.
Further, on the basis of the document selection history read from the document selection history DB 3, for each document, the occurrence probability of each document selection by the user (first statistical information) and the conditional occurrence probability of each query attribute value ( It is preferable to include a probability DB generation unit 7A that generates (second statistical information).
Furthermore, a document occurrence probability DB (database) 5 that stores the occurrence probability of the document selection generated by the probability DB generation unit 7A, and an attribute conditional occurrence probability that stores the conditional occurrence probability of each query attribute value. It is preferable to include a DB (database) 6A.

  The score calculation unit 8A obtains the occurrence probability of the document selection and the conditional occurrence probability of at least one query attribute value from the occurrence probability DBs 5 and 6A for each search result document searched by the document search unit 1. Read out and calculate the score. In this score calculation, for each search result document, a cumulative integration process of the occurrence probability of the document selection by the user and the conditional occurrence probability of at least one query attribute value is executed, and the user's current search request The score corresponding to the probability of conforming to is calculated, and the search result document list with the score or priority is returned to the document search unit 1.

Embodiment 1

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
FIG. 1 is a block diagram showing the configuration of a search result document scoring system according to the first embodiment of the present invention.
The scoring system of this embodiment is suitable for use in cooperation with a business document search system (document search unit 1), and as shown in FIG. The history DB 3 and the document scoring unit 4A are roughly configured. Here, a business is a company, organization, government office, organization, or community (hereinafter referred to as a company) composed of many departments or groups, and achieves a predetermined purpose regardless of profit or non-profit. It is a broad concept that includes the work for.

  The document search unit 1 receives a search request from a user terminal (not shown), searches for a corresponding document from search target documents, generates a search result document list, and uses the generated search result document list as a query attribute. Together with the document scoring unit 4A. When the document search unit 1 acquires a search result document list with scores from the document scoring unit 4A, the document search unit 1 rearranges the search result document list in the order of score and displays the list on the display unit of the user terminal.

  The behavior history acquisition unit 2 acquires each document displayed in the search result document list provided from the document search unit 1 together with various query attribute values such as “affiliation” and “keyword”. Here, the reaction that the user takes is, for example, an act of “selecting” a desired search result document in a broad sense from the displayed search result document list. The reaction, the reaction to download, the reaction to be registered in the favorites list, etc. are included in the user's reaction, that is, the selection action.

The document selection history DB 3 is a storage unit that accumulates user behavior information input as needed from the behavior history acquisition unit 2 in a table format as past document selection history information of the user.
Here, the user behavior information is, for example, various query attribute values, user reaction information with respect to the search result document, information on the user and the situation at the time of the reaction (for example, user name), and the like. . In the document selection history DB 3, a document selection history information table TA (FIG. 2) is set. In the document selection history information table TA, the above-described various query attribute values and user's information for the search result document are stored. Reaction information and information on the user and the situation at the time of the reaction are stored in association with each other.

  As shown in FIG. 1, the document scoring unit 4A includes a document occurrence probability DB 5, an attribute conditional occurrence probability DB 6A, a probability DB generation unit 7A, and a score calculation unit 8A. The computer is embodied by operating under the control of various programs (for example, a probability DB generation program, a score calculation program, etc.).

  The document occurrence probability DB 5 includes a document selection occurrence probability table TB (FIG. 3), and the occurrence probability of each document selection by the user as the first statistical information generated from the user's past document selection history. (Selection probability of each document) is stored in a table format. The attribute conditional occurrence probability DB 6A includes conditional occurrence probability tables TS1 and TK1 (FIG. 4) of various attributes, and each of the second statistical information generated from the user's past document selection history Stores conditional occurrence probability of query attribute value in table format. Here, the conditional occurrence probability of the query attribute value means the occurrence probability of the query attribute value for each document.

  The probability DB generation unit 7A is a data processing unit that generates / updates various statistical information (occurrence probabilities) stored in the document occurrence probability DB 5 and the attribute conditional occurrence probability DB 6A. As shown in FIG. A history acquisition unit 9a, an occurrence probability calculation unit 10a, an occurrence probability integration unit 11a, and an old occurrence probability weight calculation unit 12a are schematically configured. The document selection history acquisition unit 9a refers to the document selection history information stored in the document selection history DB 3 for each past fixed period, and determines the occurrence frequency of each document selection by the user for each document (document identifier). The conditional occurrence frequency (for each document) of each attribute value is acquired and supplied to the occurrence probability calculation unit 10a.

  The occurrence probability calculation unit 10a is based on the document selection occurrence frequency information supplied from the document selection history acquisition unit 9a, and the occurrence probability (first statistical information) of each document selection by the user over the past certain period. Is calculated. In addition, the occurrence probability calculation unit 10a determines the (document) of each query attribute value for the past certain period based on the conditional occurrence frequency (for each document) of each query attribute value supplied from the document selection history acquisition unit 9a. Every) A conditional occurrence probability (second statistical information) is calculated. Here, the query attribute (attribute value) means attribute information of the user or the search request when selecting the search result document, and the conditional occurrence probability of the query attribute value is about a certain query attribute. It is defined that the query attribute is the probability of taking a predetermined attribute value.

  The occurrence probability integration unit 11a, for each document, the occurrence probability of document selection before update (hereinafter referred to as old occurrence probability) X1a stored in the document occurrence probability DB 5 and the document selection calculated by the occurrence probability calculation unit 10a. Occurrence probability X1b (hereinafter referred to as new occurrence probability) is updated, and the old occurrence probability X1a of the document occurrence probability DB5 is updated with the integrated value X1. Further, the occurrence probability integration unit 11a has, for each document, the old conditional occurrence probability X2a of each query attribute value stored in the attribute conditional occurrence probability DB 6A and the corresponding new conditional calculation calculated by the occurrence probability calculation unit 10a. The occurrence probability X2b is integrated, and the old conditional occurrence probability X2a of the attribute conditional occurrence probability DB6A is updated with the integrated value X2.

  The weight calculation unit 12a is used when the occurrence probability integration unit 11a integrates the new and old occurrence probabilities before update (the old occurrence probability X1a of document selection and the old conditional occurrence probability X2a of each query attribute value). The weight is calculated using a function model (described later) represented by Expression (3).

As a response to the user's current search request, the score calculation unit 8A has a probability of matching the user's current search request with respect to each document (search result document) listed in the search results provided from the document search engine. Is estimated (degree of relevance to the search request), and a score corresponding to the high probability is calculated. More specifically, the score calculation unit 8A includes at least one query attribute stored in the document occurrence probability DB 5 and stored in the attribute conditional occurrence probability DB 6A. By integrating and integrating the conditional occurrence probability of the value, the probability of matching the search request is quantified as a score.
In other words, when there are a plurality of search result documents for the common query attribute value, the score calculation unit 8A calculates the score of each search result document in the common query attribute value. The priority of each search result document in the attribute value is determined.

Next, the configuration of the document selection history DB 3, the document occurrence probability DB 5, and the attribute conditional occurrence probability DB 6A will be described in detail with reference to FIGS.
First, the document selection history DB 3 will be described in detail with reference to FIG.
FIG. 2 is a conceptual diagram schematically showing an example of the document selection history information table TA set in the document selection history DB 3.
In the document selection history information table TA, as shown in FIG. 2, the user and various query attributes included in the search request when the date, user name, document identifier, and document indicated by the document identifier are selected. (Attribute values) are associated with each other and summarized in a table format.
Here, in the column item “date and time”, the occurrence date and time (for example, the date and time of occurrence) of the reaction taken by the user is described using a character string, URL, feature amount, and the like. The column item “user name” is an example of information on the user and the situation at the time of the reaction, and is set as complementary information not included in the query attribute. The user name is described using, for example, a character string (name) or a numerical value or a combination of a numerical value and a character string (employee number).

  The query attribute represents meta information for DB access included in the user and the search request. In the document selection history information table TA, as shown in FIG. 2, “affiliation” is exemplified as a query attribute possessed by a user, and “keyword” is exemplified as a query attribute possessed by a user search request. . In the query attribute “affiliation”, for example, as shown in the figure, query attribute values such as “head office”, “branch office 1”, “branch office 2” to which each user belongs are described. In the “keyword”, query attribute values such as “top page” and “entrance / exit application” are described.

  The document identifier is for specifying the document selected by the user. FIG. 2 shows an identifier consisting of a character string such as “title” or “heading” of the document, ie, “head office entry / exit application”. “Top page common to all companies” and “Application for entering and leaving branch office 2” are exemplified. As long as the user's selected document can be identified from other documents, a character string other than “title” and “heading” may be used. Alternatively, for example, a URL (Uniform Resource Locator), Alternatively, document feature quantities can be used as document identifiers.

  Based on the above, referring to the document selection history information table TA (FIG. 2), in the column of the first row, “Headquarters” requested to search with the keyword “application for entry / exit” on October 1, 2008. A record that user A belonging to A selected the document “Application for entry / exit of headquarters” at 12:35:22 while browsing the search result document list (reacted to “application for entrance / exit of headquarters”). Illustrated. Next, in the column of the second row, on October 5, 2008, the user B belonging to “Branch 1” who requested the search with the keyword “Top Page” browses the search result document list. 8:15:05, a record that a document “top page common to all companies” is selected is illustrated. Next, in the column of the third row, on October 6, 2008, the user C belonging to “Branch 2” who requested the search with the keyword “application for entry / exit” browses the search result document list. However, a record that a document of “Branch 2 Entry / Exit Application” was selected at 18:22:01 is illustrated.

Next, the document occurrence probability DB 5 will be described in detail with reference to FIG.
FIG. 3 is a conceptual diagram schematically showing an example of the document selection occurrence probability table TB set in the document occurrence probability DB 5.
In the document selection occurrence probability table TB, as shown in FIG. 3, the document identifier and the probability that the document indicated by the document identifier is selected by the user in a predetermined past period are associated with each other in a table format. It is summarized.
In this embodiment, identifiers consisting of character strings such as “title” and “headline” of the document, that is, “company-wide top page”, “head office entry / exit application”, “branch office 1 entrance / exit application”, “branch office 2 entrance” “Exit application” is exemplified as the document identifier (FIGS. 3 and 4).
Based on the above, referring to the document selection probability table TB (FIG. 3), the probability of selecting the document “Company-wide top page” is “0.5”, and the probability of selecting the document “Headquarters entrance / exit application” is It is shown that the probability of selecting “0.2”, the document “Branch 1 Entrance / Exit Application” is “0.2”, and the probability of selecting the document “Branch 2 Entrance / Exit Application” is “0.1”. Yes.

Next, the attribute conditional occurrence probability DB 6A will be described in detail with reference to FIG.
FIG. 4 is a conceptual diagram schematically showing an example of conditional occurrence probability tables TS1 and TK1 of various attributes set in the attribute conditional occurrence probability DB 6A.
In this embodiment, as the logical data structure, attribute conditional occurrence probability tables TS1 and TK1 are set independently for each query attribute. FIG. 4 illustrates the “occurrence” conditional occurrence probability table TS2 as “query attributes” possessed by the user (FIG. 4A), and also as “query attributes” possessed by the user search request. , "Keyword" conditional occurrence probability table TK2 is illustrated ((b) in the figure).
That is, in the conditional occurrence probability table TS1 of the attribute “affiliation”, as shown in FIG. 4A, the query attribute “affiliation” possessed by the user when selecting the document identifier and the document indicated by the document identifier. In the selection, the probability that the query attribute takes a predetermined attribute value is associated with each other and summarized in a table format. The query attribute “affiliation” has attribute values such as “head office”, “branch office 1”, “branch office 2” ((a) in the figure).
In addition, in the conditional occurrence probability table TK1 of the attribute “keyword”, as shown in FIG. 4B, the query attribute included in the search request of the user when selecting the document identifier and the document indicated by the document identifier. The “keyword” and the probability that the query attribute takes a predetermined attribute value at the time of selection are associated and collected in a table format. In this embodiment, the query attribute “keyword” has attribute values such as “top page” and “entrance / exit application” (FIG. 5B).

  Based on the above, referring to the conditional occurrence probability table TS1 of the attribute “affiliation” ((a) in the figure), the query attribute “affiliation” is “head office” when selecting the document “company-wide common top page”. It is shown that the probability of taking “0.497”, the probability of taking “Branch 1” is “0.256”, and the probability of taking “Branch 2” is “0.247”. In addition, when selecting the document “Headquarters entry / exit application”, the probability that the query attribute “affiliation” takes “Headquarters” is “0.700”, the probability of taking “Branch1” is “0.200”, “Branch” It is shown that the probability of taking “2” is “0.100”. Similarly, when selecting the document “Branch 1 entry / exit application”, the probability that the query attribute “affiliation” takes “head office” is “0.150”, and the probability that “branch 1” takes “0.800”, It is indicated that the probability of taking “Branch 2” is “0.050”. Similarly, when selecting the document “Branch 2 entry / exit application”, the probability that the query attribute “affiliation” takes “head office” is “0.050”, and the probability that “branch 1” takes “0.050”, It is indicated that the probability of taking “Branch 2” is “0.900”.

  Next, in the conditional occurrence probability table TK1 of the attribute “keyword”, referring to FIG. 5B, the query attribute “keyword” when selecting the document “company-wide top page” is “top page”. It is shown that the probability of taking "" is "0.90" and the probability of taking "entrance / exit application" is "0.10". In addition, when selecting the document “head office entry / exit application”, the probability that the query attribute “keyword” takes “top page” is “0.05”, and the probability of taking “entrance / exit application” is “0.95”. It is shown that there is. Similarly, when selecting the document “Branch 1 entry / exit application”, the probability that the query attribute “keyword” takes “top page” is “0.05”, and the probability of taking “entrance / exit application” is “0.95”. Is shown. Further, when selecting the document “Branch 2 entry / exit application”, the probability that the query attribute “keyword” takes “top page” is “0.10”, and the probability of taking “entrance / exit application” is “0.90”. It is shown that.

Next, the operation of the first embodiment (scoring system) will be described with reference to FIGS.
FIG. 5 is a flowchart showing the operation procedure of the score calculation process executed by the score calculation unit 8A, and FIG. 6 shows the operation procedure of the probability DB update process executed by the probability DB generation unit 7A for the various probability DBs 5 and 6A. FIG. 7 is a graph curve diagram for explaining the weight calculation process executed by the weight calculation unit 12a of the old occurrence probability constituting the same probability DB generation unit 7A.

Score Calculation Processing First, the score calculation processing will be described with reference to the flowchart of FIG. 5 and various occurrence probability tables (FIGS. 3 and 4).
First, from a user terminal (not shown), for example, a user A who works at the head office uses “head office” as the attribute value of the query attribute “affiliation” relating to the user, and the attribute value of the query attribute “keyword” relating to the document to be viewed. Enter "entrance / exit application" in sequence and make a search request to the business document search system.
When the document search unit 1 of the business document search system receives the search request from the user terminal, the document search unit 1 searches for the corresponding document from the search target documents and generates a search result document list. Here, in the search result document list, document identifiers for specifying the search result documents are listed in a number corresponding to the number of corresponding documents. In this case, it is assumed that two document identifiers (search result documents) of “company-wide common top page” and “head office entrance / exit application” are listed in this order in the generated search result document list. Next, the document search unit 1 passes the generated search result document list together with the query attribute information to the document scoring unit 4A of the scoring system, and scores the search result document (estimated relevance value of the search result). Request).

  When the score calculation unit 8A receives a scoring request from the document search unit 1 together with the search result document list and the query attribute information (step Sa0 in FIG. 5), the “company-wide common top page” and “entrance / exit to the head office” The process of sequentially calculating the score of the document identifier (search result document) “application” is started. In step Sa1, the score calculation unit 8A first refers to the document selection occurrence probability table TB (FIG. 3) set in the document occurrence probability DB 5 using the document identifier “company-wide common top page” as a search key. A value of “0.5” is acquired as an occurrence probability of selecting the search result document “company-wide common top page”.

Next, the score calculation unit 8A proceeds to step Sa2, and is set in the attribute conditional occurrence probability DB 6A using the document identifier “top page common throughout the company” and the query attribute value “head office” of “affiliation” as search keys. Reference is made to the conditional occurrence probability table TS1 (FIG. 4A) of the attribute “affiliation”. The score calculation unit 8A selects “0.497” as the conditional occurrence probability of the query attribute value “head office” that selects the document identifier “company-wide common top page” from the conditional occurrence probability table TS1 of the attribute “affiliation” to be referenced. Get the value.
Thereafter, the score calculation unit 8A uses the document identifier “company-wide common top page” and the query attribute value “entry / exit application” of “keyword” as the search keys to set the attribute “ Reference is made to the conditional occurrence probability table TK1 (FIG. 4B). The score calculation unit 8A sets “0.10” as the conditional occurrence probability of the query attribute value “entrance / exit application” in the document identifier “company-wide top page” from the conditional occurrence probability table TK1 of the attribute “keyword” to be referred to. A value is acquired (step Sa3 → Sa2 → Sa3).

  Next, the score calculation unit 8A proceeds to step Sa4, where the acquired document selection occurrence value, the conditional occurrence probability value of the query attribute value “head office”, and the query attribute value “entrance / exit application” condition The score of the search result document “company-wide common top page” is calculated by integrating and integrating the occurrence probability. In this example, the occurrence probability value “0.5” for selecting the search result document “company-wide common top page”, the conditional occurrence probability value “0.497” of the query attribute value “head office”, and the query attribute value The conditional occurrence probability value “0.10” of “entrance / exit application” is integrated and integrated [0.5 * 0.497 * 0.10]. The integrated value “0.02485” obtained in this way becomes the score of the search result document “company-wide common top page”.

When the score of the search result document “company-wide common top page” is calculated (step Sa4), the score calculation unit 8A returns to step Sa1 (step Sa5), and another search listed in the search result document list. A score calculation process is executed for the result document “head office entry / exit application”.
In step Sa1, the score calculation unit 8A selects the search result document “head office entry / exit application” with reference to the document selection probability table TB (FIG. 3) using the document identifier “head office entry / exit application” as a search key. A value of “0.2” is acquired as the occurrence probability.

  Next, the score calculation unit 8A proceeds to Step Sa2, and uses the document identifier “head office entrance / exit application” and the query attribute value “head office” of “affiliation” as search keys, and the conditional occurrence probability table TS1 of attribute “affiliation” Reference is made to FIG. The score calculation unit 8A selects “0.700” as the conditional occurrence probability of the query attribute value “head office” that selects the document identifier “head office entrance / exit application” from the conditional occurrence probability table TS1 of the attribute “affiliation” to be referenced. Get the value. Similarly, the score calculation unit 8A uses the document identifier “head office entrance / exit application” and the query attribute value “entrance / exit application” of “keyword” as search keys, and the conditional occurrence probability table TK1 (see FIG. Refer to 4 (b)). The score calculation unit 8A uses the conditional occurrence probability table TK1 of the attribute “keyword” to be referred to as “0.95” as the conditional occurrence probability of the query attribute value “entrance / exit application” with the document identifier “head office entrance / exit application”. A value is acquired (steps Sa3 and Sa2).

  Next, the score calculation unit 8A proceeds to step Sa4 again and executes the above-described integration process. In this example, the occurrence probability value “0.2” for selecting the search result document “head office entrance / exit application”, the conditional occurrence probability value “0.700” of the query attribute value “head office”, and the query attribute value The value “0.95” of the conditional occurrence probability of “entrance / exit application” is integrated and integrated [0.2 * 0.700 * 0.95]. The integrated value “0.13300” obtained in this way becomes the score of the search result document “Application for entering / exiting the head office”.

Next, the score calculation unit 8A calculates the scores for all the search result documents, and then returns the search result document list (score calculation result) with scores to the document search unit 1 (step Sa6) and ends the process. To do.
When the document search unit 1 obtains a search result document list with a score from the score calculation unit 8A, the document search unit 1 rearranges the search result document list in order of score, that is, in order of priority, and notifies the user terminal. In this example, the score of the search result document “company-wide common top page” is calculated as “0.02485”, and the score of the search result document “head office entry / exit application” is calculated as “0.13300”. Sorted in the order of “Application for entry / exit” → “Top page common to all companies” and notified to the user terminal.
In FIG. 5, processing (step Sa1) for obtaining the occurrence probability of search result document selection, processing for obtaining the conditional occurrence probability of the query attribute value (step Sa2, Sa3), which are executed by the score calculation unit 8A. Of course, the order may be changed as necessary.

Probability DB Update Processing Next, the operation of the probability DB update processing will be described with reference to the flowchart of FIG. The probability DB generation unit 7A refers to the document selection history information table TA in the document selection history DB3, and performs probability DB update processing for updating the document occurrence probability DB5 and the attribute conditional occurrence probability DB6A. .
In the probability DB generation unit 7A, the document selection history acquisition unit 9a selects from the document selection history DB 3 a recent fixed period after the old history information referenced in the probability DB update process up to the previous time (for example, the most recent month) The history information over a period of () is acquired as new history information (step Sb1 in FIG. 6). The acquired new history information is passed to the occurrence probability calculation unit 10a.

  The occurrence probability calculation unit 10a calculates a new occurrence probability X1b of document selection for each document to be searched based on the new history information given from the document selection history acquisition unit 9a, and the query attribute value for each attribute. (For each document) A new conditional occurrence probability is calculated (step Sb2). The new occurrence probability X1b of document selection calculated for each document and the new conditional occurrence probability X2b of the query attribute value calculated for each attribute are passed to the occurrence probability integration unit 11a.

The weight calculation unit 12a obtains the occurrence probability of document selection as the old occurrence probability X1a for each document to be searched from the document occurrence probability DB 5 (document occurrence probability table TB) (since it is the one before update this time). Then, the weight is calculated (step Sb3). Similarly, the weight calculation unit 12a obtains the conditional occurrence probability (for each document) of each query attribute value for each attribute from the attribute conditional occurrence probability DB 6A (attribute conditional occurrence probability table TS1, TK1). And the weight is calculated (Sb3). In this embodiment, 0 or 1 is equally assigned to all probability values as the initial setting values of the old occurrence probabilities X1a and X2a.
In this embodiment, the weight calculation unit 12a determines the query attribute value (document) of each attribute from the conditional occurrence probability tables TS1 and TK1 (FIGS. 4A and 4B) of the attributes “affiliation” and “keyword”. Every) The old conditional occurrence probability X2a is acquired and its weight is calculated. The calculated weight of the old occurrence probability X1a of document selection and the weight of the old conditional occurrence probability X2a (for each document) of the query attribute value are passed to the occurrence probability integration unit 11a.

The occurrence probability integration unit 11a calculates the new occurrence probability X1b and the old occurrence probability X1a of document selection given from the occurrence probability calculation unit 10a and the document occurrence probability DB 5 by the weight calculation unit 12a for each document to be searched. The old occurrence probability weights are used for integration (step Sb4). Similarly, the occurrence probability integration unit 11a assigns the old and new conditional occurrence probabilities X2a and X2b (for each document) of each query attribute value given from the occurrence probability calculation unit 10a and the document occurrence probability DB 5 to each attribute. Integration is performed using the weight of the occurrence probability X2a (step Sb4). In this embodiment, the integration of new and old occurrence probabilities is performed using the weighted addition method shown in equations (1) and (2).
X1 = X1a * W1 + X1b * (1-W1) (1)
X2 = X2a * W2 + X2b * (1-W2) (2)
Here, X1a is the old occurrence probability of the document selection (initial setting value is 0 or 1), X1b is the new occurrence probability of the document selection, X1 is the integrated value of the new and old occurrence probability of the document selection, and W1 is the old occurrence probability of the document selection. Is the weight. X2a is the old conditional occurrence probability of the query attribute value (initial value is 0 or 1), X2b is the new conditional occurrence probability of the query attribute value, X2 is the integrated value of the old and new conditional occurrence probability of the query attribute value, and W2 is This is the weight of the old conditional occurrence probability of the query attribute value.

  When the integration of the new and old occurrence probabilities regarding the document selection and the query attribute value is completed, the occurrence probability integration unit 11a sequentially or collectively uses the document occurrence probability DB5 or the attribute conditional occurrence probability with the integrated values X1 and X2 of the old and new occurrence probabilities. The stored contents X1a and X2a of the DB 6A are updated, and the update process is terminated (step Sb5).

Weight Calculation Processing Next, the operation of the weight calculation unit 12a will be described in further detail with reference to FIG.
When calculating the weight of the old occurrence probability, the weight calculation unit 12a increases the weight as the values of the various old occurrence probabilities X1a and X2a are lower, and decreases the weight as the value is higher, based on an algorithm incorporating a sigmoid function. (Step Sb3).
That is, for each document to be searched, the weight calculation unit 12a operates so as to increase the weight as the occurrence probability of document selection before update, that is, the old occurrence probability X1a of document selection is lower, and to decrease the weight as it is higher. . Similarly, for each document identifier for each attribute, the weight calculation unit 12a increases the weight as the old conditional occurrence probability of the query attribute value before update, that is, the old conditional occurrence probability X2a of the query attribute value decreases. It works to decrease the weight.

In this embodiment, as a weight calculation processing model of the old document occurrence probability, a function model made up of a sigmoid function that is given by Expression (3) and draws a weight curve as shown in FIG. 7 is used.
W (X; a, b, c, g) = (a−b) × S (X−c; −g) + b (3)
Here, W (X; a, b, c, g) is a weight function of the old occurrence probability, S (X; g) is a sigmoid function expressed by the equation (4), X is an old occurrence probability, and g is a sigmoid. The gain of the function, c is the X value of the inflection point of the sigmoid curve, a is a preset maximum value, and b is a preset minimum value.
S (X; g) = (1 + e ^−gx ) ⁻¹ (4)

The weight W of the old occurrence probability of document selection is obtained by substituting the value of the old occurrence probability into the variable X of the function model given by Expression (3). For example, when calculating the weight of the old occurrence probability with respect to the selection of the document “head office entry / exit application”, the weight calculation unit 12a generates the document selection occurrence probability table TB set in the document occurrence probability DB 5 (FIG. 3). ) To obtain the old occurrence probability “0.2” of the “head office entry / exit application” selection. Next, the weight calculation unit 12a performs arithmetic processing by substituting the acquired old occurrence probability “0.2” into the variable X of the function model of Expression (3), and sets the weight function W as [0. 7] is calculated. From this calculation result, the weight calculation unit 12a assigns “0.7” as the weight of the old occurrence probability of the “head office entry / exit application” selection.
Similarly, for example, regarding the selection of the document “company-wide common top page”, when calculating the weight of the old occurrence probability, the weight calculation unit 12a refers to the document selection occurrence probability table TB, Acquire old occurrence probability “0.5” of “page” selection. Next, the weight calculation unit 12a performs an arithmetic process by substituting the acquired old occurrence probability “0.5” into the variable X of the function model of Expression (3), and the value of the weight function W is [0. 23] is calculated. From this calculation result, the weight calculation unit 12a assigns “0.23” as the weight of the old occurrence probability of selecting the “company-wide common top page”.
As described above, when calculating the weight of the old occurrence probability, the weight calculation unit 12a increases the weight for the “head office entrance / exit application” having a low value of the old occurrence probability of document selection, For the “company-wide common top page” having a high occurrence probability value, the weight is reduced (step Sb3).

The weight W of the old conditional occurrence probability of the query attribute value can also be obtained by substituting the old occurrence probability value into the variable X of the function model given by Expression (3).
In consideration of the difference in properties between the selected document and the query attribute, even when the same function model is used, the various parameters a, b, c, and g are used to calculate the weight of the old occurrence probability of the document selection and the query. Different values may be taken for the weight calculation of the old conditional occurrence probability of the attribute value. The parameters can be increased or decreased as necessary. Similarly, considering the difference in properties from various “query attributes”, the various parameters a, b, c, and g have different values for each attribute even when the function model of Expression (3) is used in common. In this case as well, the parameters can be increased or decreased as necessary.

As described above, according to the above configuration, the score calculation unit 8A refers to the conditional occurrence probability of the query attribute value stored for each query attribute, and determines the occurrence probability of document selection and the occurrence probability of various query attribute values. Is integrated and integrated.
For this reason, even if the relevance relevance of the search result document changes due to a change in business due to a change in the user's affiliation or job title, even if the relevance relevance of the search result document changes, an appropriate score that accurately follows the relevance change Ring can be done.
In addition, according to the above configuration, even if the community system partially changes, if there is no change in other attributes, for example, the user's business or work location, etc. , It is possible to avoid the influence of fluctuation factors as much as possible. Therefore, appropriate scoring can be continuously performed as a whole.

In addition, when the value of the old occurrence probability before update is high, the weight calculation unit 12a executes a process of reducing the weight of the old occurrence probability. In the occurrence probability integration unit 11a using the weights thus obtained, the old and new occurrence probabilities are integrated, and the document occurrence probability DB 5 and the attribute conditional occurrence probability DB 6A are updated with the integrated value of the old and new occurrence probabilities. Such a configuration can prevent a nonconforming document from being ranked higher in the search result document list.
In general, non-conforming documents (noise) tend to decrease afterward because of the learning ability of the user even if the selection probability rises temporarily. However, according to this embodiment, a low weight is assigned to the old occurrence probability of the nonconforming document selection that temporarily increases. Also, thanks to the learning ability of the user, the probability of new occurrence of nonconforming document selection is considered to be low. Therefore, the selection probability of the nonconforming document is larger than the simple weighted weighting, and therefore the selection probability of the nonconforming document is relatively greatly reduced with respect to the selection probability of the conforming document. The ranking will drop rapidly. As a result, the ranking of relevant documents rises quickly. Therefore, even when a nonconforming document is included in the search result document list, the influence can be suppressed small.

Embodiment 2

Next, a second embodiment will be described with reference to FIG.
FIG. 8 is a block diagram showing the configuration of a search result document scoring system according to the second embodiment of the present invention.
In the second embodiment, there is a statistically significant difference between the distribution of the conditional occurrence probability (for each document) of the query attribute value and the distribution of the occurrence probability of the query attribute value (regardless of the document). This is different from that of the first embodiment in that score calculation is performed using different probability sources based on the determination result.
In order to implement the calculation method of this embodiment, as shown in FIG. 8, the document scoring unit 4B includes an attribute occurrence probability DB (database) 13B for storing third statistical information to be described later and a feature detection filter unit. 14B is added. In FIG. 8, the same parts (or corresponding parts) as those shown in FIG. 1 are denoted by the same reference numerals (or the same reference numerals), and the description thereof is omitted or simplified.

  The document selection history acquisition unit 9b acquires the occurrence frequency of each document selection and the conditional occurrence frequency (for each document) of each query attribute value from the document selection history DB 3 for each fixed period in the past, and for the selected document. Regardless, the occurrence frequency of each query attribute value is acquired and supplied to the occurrence probability calculation unit 10b. The occurrence probability calculation unit 10b calculates the occurrence probability (first statistical information) of each document selection by the user over the past fixed period based on the supplied document selection occurrence frequency information. Also, the occurrence probability calculation unit 10b determines the conditional occurrence probability (second statistic) of each query attribute value (for each document) over the past certain period based on the conditional occurrence frequency of each supplied query attribute value. Information). Furthermore, the occurrence probability calculation unit 10b calculates the occurrence probability (third statistical information) of each query attribute value over the past fixed period based on the occurrence frequency of each supplied query attribute value.

The occurrence probability integration unit 11b uses, for each document, the old occurrence probability X1a of the document selection before update stored in the document occurrence probability DB 5 and the new occurrence probability X1b of the document selection calculated by the occurrence probability calculation unit 10b. The old occurrence probability X1a of the document occurrence probability DB5 is updated with the integration value X1. Further, the occurrence probability integration unit 11a, for each document, the old conditional occurrence probability X2a of each query attribute value stored in the attribute conditional occurrence probability DB 6B and the corresponding new condition calculated by the occurrence probability calculation unit 10. The added occurrence probability X2b is integrated, and the stored content X2a of the attribute conditional occurrence probability DB 6B is updated with the integrated value X2. Further, the occurrence probability integration unit 11a integrates, for each document, the old occurrence probability X3a of each query attribute value stored in the attribute occurrence probability DB 13B and the corresponding new occurrence probability X3b calculated by the occurrence probability calculation unit 10b. Then, the stored content X3a of the attribute occurrence probability DB 13B is updated with the integrated value X3.
The weight calculation unit 12b is used when the occurrence probability integration unit 11b integrates the old and new occurrence probabilities, the old occurrence probabilities before update (the old occurrence probability X1a of document selection, the old conditional occurrence probability X2a of each query attribute value, and the occurrence The weight of the probability X3a) is calculated using a function model (for example, equation (3)).

The attribute occurrence probability DB 13B includes occurrence probability tables TS2 and TK2 (FIGS. 9A and 9B) of various attributes, and is used as third statistical information generated from the user's past document selection history. The occurrence probability of each query attribute value for each attribute is stored in a table format.
Here, the occurrence probability of the query attribute value means the probability that each attribute value takes for a certain attribute regardless of the selected document, and is the second statistical information of the query attribute value (for each document). This is different from the conditional occurrence probability (FIG. 4).

When updating the DB, the feature detection filter unit 14B uses the integrated value X2 of the new and old conditional occurrence probabilities of the query attribute values generated by the occurrence probability integration unit 11b for each attribute, and stores the storage content X2a of the attribute conditional occurrence probability DB 6B. Updates are permitted or prohibited on a document-by-document basis.
That is, the feature detection filter unit 14B, when updating the DB, the distribution of the conditional occurrence probability (integrated value) X2 of the query attribute value for each attribute newly generated by the occurrence probability integration unit 11b, and the query attribute The distribution of the value occurrence probability (integrated value) Y2 is compared. When there is a statistically significant difference between the two probability distributions as a result of the comparison, the feature detection filter unit 14B generates an attribute condition with the conditional occurrence probability (integrated value) X2 of the generated query attribute value. The update of the stored content X2a of the probability DB 6B is permitted. Here, “there is a statistically significant difference between the two probability distributions” means that the probability distribution of the document selection “has a characteristic distribution” for the attribute.

  On the other hand, if there is no statistically significant difference between the two probability distributions as a result of the comparison, the feature detection filter unit 14B writes the conditional occurrence probability X2 of the query attribute value to the attribute conditional occurrence probability DB 6B. Is prohibited. “There is no statistically significant difference between the two probability distributions” means that the probability distribution of the document selection “has no characteristic distribution” for the attribute. The feature detection filter unit 14B replaces the writing of the conditional occurrence probability X2 with the information (described later) indicating that the probability distribution of the document selection “has no characteristic distribution” for the attribute. Write to the conditional occurrence probability DB 6B.

Next, the attribute occurrence probability DB 13B will be described in detail with reference to FIG.
FIG. 9 is a conceptual diagram schematically showing an example of the occurrence probability tables TS2 and TK2 of various attributes set in the attribute occurrence probability DB 13B. In this embodiment, the logical data structure is as follows for each query attribute. Independently, attribute generation probability tables TS2 and TK2 are set. FIG. 9 illustrates an occurrence probability table TS2 of “affiliation” as a “query attribute” possessed by a user (FIG. 9A), and “query attribute” possessed by a user search request includes “ The occurrence probability table TK2 of “keyword” is illustrated ((b) in the figure).

That is, in the occurrence probability table TS2 of the attribute “affiliation”, as shown in FIG. 9A, the query attribute “affiliation” possessed by the user when selecting any document (not a specific document) is shown. In the selection, the probability that the query attribute takes a predetermined attribute value is associated with each other and summarized in a table format. The query attribute “affiliation” has attribute values such as “head office”, “branch office 1”, “branch office 2” ((a) in the figure).
In addition, in the occurrence probability table TK1 of the attribute “keyword”, as shown in FIG. 9B, the query attribute “having the search request of the user when selecting any document (not a specific document)” The “keyword” and the probability that the query attribute takes a predetermined attribute value at the time of selection are associated with each other and collected in a table format. In this embodiment, the query attribute “keyword” has attribute values such as “top page” and “entrance / exit request” ((b) in the figure).

Based on the above, referring to the occurrence probability table TS2 of the attribute “affiliation” ((a) in the figure), regardless of the selected document, the probability that the query attribute “affiliation” takes “head office” is “0.50”. The probability of taking “Branch 1” is “0.250” and the probability of taking “Branch 2” is “0.250”.
Next, in the occurrence probability table TK2 of the attribute “keyword”, referring to FIG. 5B, the probability that the query attribute “keyword” takes “top page” is “0” regardless of the selected document. .20 ”, the probability of taking“ application for entrance / exit ”is“ 0.80 ”.

Next, the attribute conditional occurrence probability DB 6B will be described in detail with reference to FIG.
FIG. 10 is a conceptual diagram schematically showing an example of conditional occurrence probability tables TS3 and TK3 for various attributes set in the attribute conditional occurrence probability DB 6B.
In the conditional occurrence probability tables TS3 and TK3 of the attribute, as shown in FIGS. 9A and 9B, the probability distribution of the document selection indicates that the attribute has “no characteristic distribution”. A column for setting a flag indicating the information to be shown is added. The flag is described by, for example, a numerical value indicating true / false, a Boolean algebra, a character string (TRUE) indicating true / false, and the like.

  In this embodiment, when the flag “TRUE” is set, the probability distribution of the document selection indicates that the attribute has “no characteristic distribution”. In the example of FIG. 10A, since the flag “TRUE” is set in the row of the document identifier “company-wide common top page”, the probability distribution of the document selection “company-wide common top page” has the attribute “affiliation”. For “has no characteristic distribution”. As shown in FIG. 6A, the document identifier with the flag “TRUE” is not provided with a row for each query attribute value, and the description of the probability value for each query attribute value is prohibited from being deleted. . This is because the probability distribution of document selection is reflected for the attribute, and for the document “having no characteristic distribution”, the distribution is reflected by referring to the attribute occurrence probability tables TS2 and TK2. This is because it is not necessary to make redundant descriptions in the conditional occurrence probability tables TS3 and TK3.

  The score calculation unit 8B refers to the attribute conditional occurrence probability tables TS3 and TK3, and performs the same score calculation as described in the first embodiment when the flag of “TRUE” is not set. On the other hand, when the flag “TRUE” is set as a result of the reference, the score calculation unit 8B does not describe the conditional occurrence probability of the attribute in the attribute conditional occurrence probability tables TS3 and TK3. Therefore, instead, the score is calculated using the occurrence probability of the query attribute value stored in the attribute occurrence probability DB 13B.

Next, the operation of the second embodiment will be described with reference to FIGS.
FIG. 11 is a flowchart showing the operation procedure of the score calculation process executed by the score calculation unit 8B. FIG. 12 shows the probability DB update process executed by the probability DB generation unit 7B on the various probability DBs 5, 6B and 13B. It is a flowchart which shows an operation | movement procedure.

Score Calculation Processing First, the operation of the score calculation processing in the second embodiment will be described with reference to the flowchart of FIG. 11 and various occurrence probability tables (FIGS. 3, 9, and 10).
First, from a user terminal (not shown), for example, a user A who works at the head office uses “head office” as the attribute value of the query attribute “affiliation” relating to the user, and the attribute value of the query attribute “keyword” relating to the document to be viewed. Enter "entrance / exit application" in sequence and make a search request to the business document search system.
When the document search unit 1 of the business document search system receives the search request from the user terminal, the document search unit 1 searches for the corresponding document from the search target documents and generates a search result document list. In this case, it is assumed that two document identifiers (search result documents) of “company-wide common top page” and “head office entrance / exit application” are listed in this order in the generated search result document list. Next, the document search unit 1 passes the generated search result document list together with the query attribute information to the document scoring unit 4B of the scoring system, and requests scoring for the search result document. In the following description, items that are simplified or omitted are substantially the same as those described in the first embodiment.

  When the score calculation unit 8B receives a scoring request from the document search unit 1 together with the search result document list and the query attribute information (step Pa0 in FIG. 11), the “company-wide common top page” and “entrance / exit to the head office” The process of sequentially calculating the score of the document identifier (search result document) “application” is started. In Step Pa1, the score calculation unit 8B first refers to the document selection occurrence probability table TB (FIG. 3) set in the document occurrence probability DB 5 using the document identifier “top page common throughout the company” as a search key. A value of “0.5” is acquired as an occurrence probability of selecting the search result document “company-wide common top page”.

  Next, the score calculation unit 8B proceeds to step Pa2, and is set in the attribute conditional occurrence probability DB 6B using the document identifier “top page common throughout the company” and the query attribute value “head office” of “affiliation” as search keys. Reference is made to the conditional occurrence probability table TS3 (FIG. 10A) of the attribute “affiliation”. The score calculation unit 8B refers to the column item of the flag in the row of the document identifier “company-wide common top page”, and the probability distribution of the “company-wide common top page” selection selects “characteristic distribution” for the attribute “affiliation”. It is determined whether or not it has. In FIG. 10A, since the flag “TRUE” is set in the row of the document identifier “company-wide common top page”, the score calculation unit 8B indicates that the probability distribution of the document “company-wide common top page” selection is For the attribute “affiliation”, a determination result of “having no characteristic distribution” is obtained. When the score calculation unit 8B obtains a determination result of “having no characteristic distribution”, the process proceeds to step Pa4.

  In step Pa4, the score calculation unit 8B generates the attribute “affiliation” set in the attribute occurrence probability DB 13B using the document identifier “top page common throughout the company” and the query attribute value “head office” of “affiliation” as search keys. With reference to the probability table TS2 (FIG. 9A), a value of “0.5” is acquired as the probability of occurrence of selecting the search result document “company-wide common top page”.

After that, the score calculation unit 8B returns to step Pa2, and uses the document identifier “company-wide common top page” and the query attribute value “entrance / exit application” of the attribute “keyword” as a search key and sets the condition of the attribute “keyword”. Refer to the occurrence probability table TK3 (FIG. 10B). The score calculation unit 8B refers to the column item of the flag of the row of the document identifier “company-wide common top page”, and the probability distribution of the “company-wide common top page” selection selects “characteristic distribution” for the attribute “keyword”. It is determined whether or not it has.
In FIG. 10B, since the flag of “TRUE” is not set in the row of the document identifier “company-wide common top page”, the score calculation unit 8B has the probability distribution of selecting the document “company-wide common top page” as follows: A determination result of “having a characteristic distribution” is obtained for the attribute “keyword”. When the score calculation unit 8B obtains the determination result “having a characteristic distribution”, the process proceeds to step Pa4. In step Pa4, the row item of the document identifier “company-wide common top page” of the conditional occurrence probability table TK3 of the attribute “keyword” is referred to, and the query attribute value “entrance / exit application” is found in the document identifier “company-wide common top page”. A value of “0.10” is acquired as the conditional occurrence probability (steps Pa5 → Pa2 → Pa3 → Pa5).

  Next, the score calculation unit 8B proceeds to step Pa6, where the acquired document selection occurrence probability value “0.5”, the query attribute value “head office” occurrence probability value “0.5”, and the query attribute The value “0.1” of the conditional occurrence probability of the value “application for entrance / exit” is integrated and integrated [0.5 * 0.5 * 0.10], and the score “ 0.025 "is calculated.

  When the score of the search result document “company-wide common top page” is calculated (step Pa6), the score calculation unit 8B returns to step Pa1 (step Pa7), and another search listed in the search result document list. The above score calculation process (steps Pa1 to Pa6) is repeated for the result document “head office entry / exit application”.

Next, after calculating the scores for all the search result documents, the score calculation unit 8B returns the search result document list (score calculation result) with scores to the document search unit 1 (step Pa8) and ends the processing. To do.
When the document search unit 1 acquires the search result document list with scores from the score calculation unit 8B, the document search unit 1 rearranges the search result document list in order of score, that is, in order of priority, and notifies the user terminal.

Probability DB Update Processing Next, the operation of the probability DB update processing in the second embodiment will be described with reference to the flowchart of FIG.
The probability DB generation unit 7B refers to the document selection history information table TA in the document selection history DB3, and updates the document occurrence probability DB5, the attribute conditional occurrence probability DB6B, and the attribute occurrence probability DB13B. Perform the update process. In the probability DB generation unit 7B, the document selection history acquisition unit 9b selects a recent fixed period after the old history information referred to in the probability DB update process up to the previous time from the document selection history DB 3 (for example, the most recent month) The history information over a period of () is acquired as new history information (step Pb1 in FIG. 12). The acquired new history information is passed to the occurrence probability calculation unit 10b.

  The occurrence probability calculation unit 10b calculates a new occurrence probability X1b of document selection for each search target document based on the new history information given from the document selection history acquisition unit 9b, and the query attribute value for each attribute. (For each document) A new conditional occurrence probability and a new occurrence probability X3b (regardless of the document) are calculated (step Pb2). The new occurrence probability X1b of document selection calculated for each document, the new conditional occurrence probability X2b and the new occurrence probability X3b of the query attribute value calculated for each attribute are passed to the occurrence probability integration unit 11b.

  The weight calculation unit 12b obtains the old occurrence probability X1a of the document selection for each document to be searched from the document occurrence probability DB5 (document selection occurrence probability table TB), and calculates the weight (Step Pb3). Similarly, the weight calculation unit 12b acquires the old conditional occurrence probability X2a (for each document) of each query attribute value for each attribute from the attribute conditional occurrence probability DB 6B (attribute conditional occurrence probability table TS3, TK3). Then, the weight is calculated (Pb3). Further, the weight calculation unit 12b acquires the old occurrence probability X3a (regardless of the document) of each query attribute value for each attribute from the attribute occurrence probability DB 13B (attribute occurrence probability table TS2, TK2). The weight is calculated (same Pb3). The calculated weight of the old occurrence probability X1a of document selection and the old conditional occurrence probability X2a and old occurrence probability X3a of the query attribute value are passed to the occurrence probability integration unit 11b.

  For each document, the occurrence probability integration unit 11b uses the old and new occurrence probabilities X1b and X1a of document selection given from the occurrence probability calculation unit 10b and the document occurrence probability DB 5, and the weight of the old occurrence probability calculated by the weight calculation unit 12b. And integrated (step Pb4). Similarly, the occurrence probability integration unit 11b uses, for each attribute, the new and old conditional occurrence probabilities X2a and X2b of the query attribute values given from the occurrence probability calculation unit 10a and the document occurrence probability DB 5 as the old conditional occurrence probability X2a. Integration is performed using weights (step Pb4). Further, the occurrence probability integration unit 11b generates old and new occurrence probabilities X3a and X3b (regardless of the document) of each query attribute value given from the occurrence probability calculation unit 10a and the attribute occurrence probability DB 13B for each attribute. Integration is performed using the weight of the probability X3a (step Pb4).

In this embodiment, the new and old occurrence probabilities are integrated by using the weighted addition method shown in Expression (5) in addition to Expressions (1) and (2) similar to those described in the first embodiment.
X1 = X1a * W1 + X1b * (1-W1) (1)
X2 = X2a * W2 + X2b * (1-W2) (2)
X3 = X3a * W3 + X3b * (1-W3) (3)
Here, X1a is the old occurrence probability of the document selection (initial setting value is 0 or 1), X1b is the new occurrence probability of the document selection, X1 is the integrated value of the new and old occurrence probability of the document selection, and W1 is the old occurrence probability of the document selection. Is the weight. X2a is the old conditional occurrence probability of the query attribute value (initial value is 0 or 1), X2b is the new conditional occurrence probability of the query attribute value, X2 is the integrated value of the old and new conditional occurrence probability of the query attribute value, and W2 is This is the weight of the old conditional occurrence probability of the query attribute value. X3a is the old occurrence probability of the query attribute value (initial value is 0 or 1), X3b is the new occurrence probability of the query attribute value, X3 is the integrated value of the old and new occurrence probability of the query attribute value, and W3 is the old occurrence of the query attribute value Probability weight.

  When the integration of the new and old occurrence probabilities regarding the document selection and the query attribute value is completed, the occurrence probability integration unit 11b sequentially or collectively uses the integrated values X1 and X3 of the old and new occurrence probabilities, and the document occurrence probability DB5 or the attribute occurrence probability DB 13B. The stored contents X1a and X3a are updated (step Pb4). When the integration of the new and old occurrence probabilities and the new and old conditional occurrence probabilities for the query attributes is completed, the occurrence probability integration unit 11b passes the integrated values X2 and X3 of the new and old occurrence probabilities to the feature detection filter unit 14B (step Pb4).

  The feature detection filter unit 14B compares the probability distribution between the integrated value X2 of the new and old conditional occurrence probability of the query attribute value given from the occurrence probability integration unit 11b and the integrated value X3 of the old and new occurrence probability, and there is a significant difference. Is determined to be “characteristic distribution” (step Pb5). When the feature detection filter unit 14B determines that the distribution is “characteristic distribution” as a result of the comparison, the conditional occurrence probability (integrated value) X2 of the attribute having the characteristic distribution is used as the attribute conditional occurrence probability DB 6B. It records in (attribute conditional occurrence probability table TS3, TK3). On the other hand, when there is no significant difference between the two occurrence probabilities (integrated values) X2 and X3 as a result of the comparison, the feature detection filter unit 14B stores the attribute for the document in the attribute conditional occurrence probability tables TS3 and TK3. Instead of recording the conditional occurrence probability (integrated value) X2, a flag of “TRUE” indicating “no characteristic distribution” is set (FIG. 10A).

  That is, in this embodiment, only the conditional occurrence probability “having a characteristic distribution” for the attribute is recorded in the attribute conditional occurrence probability DB 6B, and the document selection probability distribution is “having a characteristic distribution”. The “none” conditional occurrence probability is not recorded in the attribute conditional occurrence probability DB 6B. Since it can be assumed that the conditional occurrence probability “without characteristic distribution” for the attribute is collectively recorded in the attribute occurrence probability DB 13B (attribute occurrence probability table TS2, TK2), the attribute conditional occurrence probability DB 6B This is because the recording to can be omitted.

Here, what can be said to be statistically not significant can be arbitrarily defined as necessary. For example, a test may be used. In this embodiment, for the sake of simplicity, the difference between the conditional occurrence probability (integrated value) X2 and the occurrence probability (integrated value) X3 is within a predetermined small range for all query attribute values constituting the predetermined attribute. (For example, within “0.05”), it is determined that there is no statistically significant difference.
For example, when the document identifier “company-wide common top page” is selected, the conditional occurrence probability (integrated value) X2 of the query attribute “affiliation” takes “head office” as the query attribute value “0.497”, “ It is assumed that the distribution is performed at a rate of “0.256” for taking “branch 1” and “0.247” for taking “branch 2”.

  On the other hand, the occurrence probability (integrated value) X3 of the query attribute value “affiliation” recorded in the attribute occurrence probability table TS2 is, as shown in FIG. .500 ”, the probability of taking“ Branch 1 ”is“ 0.250 ”, and the probability of taking“ Branch 2 ”is“ 0.250 ”.

In the case of such a probability distribution, for all query attribute values constituting the attribute “affiliation”, the difference between the conditional occurrence probability (integrated value) X2 and the occurrence probability (integrated value) X3 is the largest, Since it is “0.006”, it is determined that there is no statistically significant difference. Therefore, the feature detection filter unit 14B determines that the probability distribution of selecting the document “company-wide common top page” is “has no characteristic distribution” for the attribute “affiliation” (step Pb5).
Therefore, the feature detection filter unit 14B records the conditional occurrence probability (integrated value) X2 of the query attribute value when selecting the document “company-wide common top page” in the conditional occurrence probability table TS3 of the attribute “affiliation”. Instead, a flag of “TRUE” indicating that “there is no characteristic distribution” is set in the line item of the document “top page common throughout the company” (FIG. 10A). By setting a flag column in the conditional occurrence probability tables TS3 and TK3 (FIG. 10), in the conditional occurrence probability tables TS1 and TK1 (FIG. 4) used in the first embodiment, the number of query attribute values However, the number of rows of the “company-wide common top page” is required, but in the second embodiment, it can be omitted.

  When the feature detection filter unit 14B completes the recording process / flag setting process based on the comparison determination described above for all documents and all query attributes, the probability DB generation unit 7B ends the DB update process (Step S1). Sb6).

  Therefore, according to the second embodiment, substantially the same effect as described in the first embodiment can be obtained. In addition, according to the second embodiment, conditional occurrence probability information of attributes having no characteristic distribution can be removed from the attribute conditional occurrence probability DB 6B as redundant data, so that necessary storage device resources, Calculation resources can be reduced.

Modified Example FIG. 13 is a block diagram showing a configuration of a search result document scoring system which is a modified example of the second embodiment.
In the second embodiment (FIG. 8), the feature detection filter unit 14B acquires the attribute occurrence probability X3 from the occurrence probability integration unit 11b. However, the present invention is not limited to this, as shown in FIG. The detection filter unit 14C may acquire the attribute occurrence probability X3 directly from the attribute occurrence probability DB 13C. In this case, it is preferable that the occurrence probability integration unit 11c updates the storage content X2a of the attribute occurrence probability DB 13 before passing the conditional occurrence probability of the attribute to the feature detection filter unit 14C. In FIG. 13, the same parts (or corresponding parts) as those shown in FIG. 8 are denoted by the same reference numerals (or the same reference numerals), and the description thereof is omitted or simplified.

As described above, the embodiment of the present invention has been described in detail with reference to the drawings. However, the specific configuration is not limited to this embodiment, and there are design changes and the like within the scope not departing from the gist of the present invention. However, it is included in this invention. For example, in the above-described embodiment, the score is calculated by simply integrating and integrating various occurrence probabilities, but the present invention is not limited to this. For example, after weighting various occurrence probabilities The integration integration process may be performed. In the above-described embodiment, the new and old occurrence probabilities are integrated using the weighted addition method. However, the present invention is not limited to this, and weighting may be omitted as necessary.
In addition, the sigmoid function is used as the weight calculation processing model. However, the present invention is not limited to this, and a function model composed of a linear function, a quadratic function, a trigonometric function, another exponential function, or a combination thereof is used as necessary. Also good. In addition, different function models and combinations of different parameters may be used for each type of occurrence probability handled.

In the first embodiment described above, as shown in FIG. 3, the document occurrence probability DB 5 has been described as being composed of a single table (document selection occurrence probability table TB). If it has a data structure equivalent to that shown in FIG. 3, it may be composed of a plurality of tables.
In the document search unit 1 described above, the search result document list is rearranged in order of score based only on the score acquired from the document scoring unit 4A. However, the present invention is not limited to this. For example, a score such as TF (Term Frequency) -IDF (Inverse Document Frequency) generated by the document search unit 1 itself and a score generated by the document scoring unit 4A are integrated, and based on the score obtained by the integration. The search result document list may be rearranged.
In the first embodiment, as shown in FIG. 4, a conditional occurrence probability table for each query attribute is provided. However, the present invention is not limited to this, and can be logically decomposed into a table for each attribute. If there is, the attribute conditional occurrence probability DB 6A may be composed of a table including a plurality of types of query attributes as items. On the contrary, as long as it can be logically integrated into a table for each query attribute, the attribute conditional occurrence probability DB 6A may be composed of a larger number of tables than the number of types of query attributes.

  In each of the above-described embodiments, two types of attributes of “affiliation” and “keyword” are used as the sample space of the conditional occurrence probability necessary for calculating the score. Not limited to two types, it can be increased or decreased as necessary. For example, the larger the business scale of the company to which the user belongs, and the more the type of business is distributed, the more the number of query attributes as a sample space may be increased. If the business scale of a company is small and the type of industry is concentrated, only one query attribute as a sample space may be used.

  In each of the above-described embodiments, “affiliation” and “keyword” are used as query attributes. However, the query attributes are not limited to these examples, and different attributes are used singly or in any combination. Of course, it may be. Examples of other types of attributes include “position”, “year of service”, “work location”, “search date”, “search date”, and the like. Note that continuous information such as “search date and time” and “search date” may be divided and held in time zones or periods having arbitrary width values.

  In the second embodiment described above, the column item “flag” is provided in the conditional occurrence probability table of the attribute in order to indicate that the probability distribution of the document selection is not a characteristic distribution for the attribute. Not limited to this. Instead of the flag, for example, by specifying a null character (null) in the attribute or probability column, or by setting a dedicated table indicating “the presence or absence of a characteristic probability distribution of document selection” for each attribute, A function similar to that of the flag can also be provided.

  The present invention can be applied to an information retrieval apparatus that retrieves related documents from a document index. It can be used not only within one company but also in a wide community or between communities.

1 Document search part (document search means)
2 Action history acquisition unit 3 Document selection history DB (document selection history storage unit)
TA document selection history information table 4A, 4B, 4C Document scoring unit 5 Document occurrence probability DB (probability information storage unit)
TB document selection occurrence probability table 6A, 6B Occurrence probability DB with attribute condition (probability information storage unit)
TS1, TS3 Conditional occurrence probability table with attribute “affiliation” TK1, TK3 Conditional occurrence probability table with attribute “keyword” 7A, 7B, 7C Probability DB generation unit (probability information generation means)
8A, 8B Score calculation unit (score calculation means)
9a, 9b Document selection history acquisition unit (part of probability information generation means)
10a, 10b Occurrence probability calculation unit (part of probability information generation means)
11a, 11b, 11c Occurrence probability integration unit (part of probability information generation means)
12a, 12b Weight calculation unit (part of probability information generation means)
13B, 13C Attribute occurrence probability DB (probability information storage unit)
TS2 attribute “affiliation” occurrence probability table TK2 attribute “keyword” occurrence probability table 14B, 14C feature detection filter unit (filter means)

Claims (38)

As a response to the user's current search request, for each document listed in the search results provided by the document search means,
The probability of occurrence of the document selection by the user as the first statistical information respectively generated from the past document selection history of the user, and the current time when selecting the document as the second statistical information Based on the conditional occurrence probability of at least one attribute information that the user or the search request has,
A scoring system for search result documents, comprising score calculation means for estimating a probability of conformity to a user's current search request and digitizing the score. When selecting the document as the first statistical information as the first statistical information and the document as the second statistical information from the user's past document selection history as the first statistical information Probability information generation means for generating a conditional occurrence probability of each attribute information that the user or the search request has,
As a response to the user's current search request, the user as the first statistical information generated by the probability information generation unit for each document listed in the search result provided by the document search unit On the basis of the occurrence probability of the document selection by and the conditional occurrence probability of at least one attribute information possessed by the current user or the search request when selecting the document as the second statistical information, A scoring system for search result documents, comprising score calculation means for estimating a probability of conformity to a user's current search request and digitizing the score.   The score calculation means, for each document listed in the search result based on the user's current search request, as the first statistical information, the occurrence probability of the document selection, and the second statistical information 3. The search result document scoring system according to claim 1, wherein the score is calculated by integrating and integrating the conditional occurrence probability of at least one of the attribute information.   The conditional occurrence probability of the attribute information, as the second statistical information, is calculated for each document to be searched and for at least one attribute of the user or the search request when selecting the document. 3. The scoring system for search result documents according to claim 1, wherein the attribute is defined as a probability of taking each attribute value. The occurrence probability of the document selection as the first statistical information and the conditional occurrence probability of each attribute information as the second statistical information, which are respectively generated by the probability information generation means for each search target document And a probability information storage unit for storing
3. The scoring of a search result document according to claim 2, wherein the score calculation means calculates the score by reading out the corresponding first and second statistical information from the probability information storage unit. system. A document selection history storage unit for storing a user's past document selection history based on the search result provided from the document search means; and
3. The scoring of a search result document according to claim 2, wherein the probability information generation means generates the first and second statistical information by reading the document selection history from the document selection history storage unit. system. The probability information generating means includes
From the user's past document selection history, for each search target document, the frequency of occurrence of the document selection by the user, and the condition of each attribute information that the user or the search request has when selecting the document And the frequency of occurrence
For each document to be searched, the occurrence probability of the document selection as the first statistical information is generated based on the obtained occurrence frequency of the document selection, and the conditional occurrence frequency of each acquired attribute information is set. 3. The scoring system for search result documents according to claim 2, wherein a conditional occurrence probability of each attribute information as the second statistical information is generated based on. The probability information generating means includes
For each document to be searched from the past document selection history of the user, the previous occurrence probability of the document selection calculated based on the occurrence frequency of the document selection acquired in the previous past period, and the previous past period Integrating the new occurrence probability of the document selection calculated based on the frequency of occurrence of the document selection acquired in step (i), generating and updating the occurrence probability of the document selection as the first statistical information, and
For each document to be searched, the old conditional occurrence probability of each attribute information calculated based on the conditional occurrence frequency of each attribute information acquired in the previous past period, and each of the above acquired in the previous previous period The conditional occurrence probability of each attribute information as the second statistical information is generated and updated by integrating the new conditional occurrence probability of each attribute information calculated based on the conditional occurrence frequency of the attribute information The scoring system for search result documents according to claim 2. The probability information generating means includes
When generating and updating the occurrence probability of the document selection and the conditional occurrence probability of each attribute information, at least weight the old occurrence probability of the document selection and the old conditional occurrence probability of each attribute information ,
The weight of the old occurrence probability of the document selection or the old conditional occurrence probability of each attribute information is set to be smaller as the value of the old occurrence probability of the document selection or the old conditional occurrence probability of each attribute information is larger. The scoring system for search result documents according to claim 8. The probability information generation means reads the document selection history from the document selection history storage unit, and in addition to the first and second statistical information, a user at the time of a search request as third statistical information Or, generate the probability of occurrence of attribute information that the search request has,
The probability information storage unit stores the third statistical information generated by the probability information generation unit, and
When calculating the score for a certain document, the score calculating means generates a conditional occurrence probability distribution of attribute information as the second statistical information and generation of attribute information as the third statistical information. If there is no statistically significant difference in the probability distribution,
The score is calculated based on an occurrence probability of the document selection and the third statistical information as the first statistical information instead of the second statistical information. 2. A scoring system for search result documents according to 2.   The probability of occurrence of the attribute information as the third statistical information is defined as the probability that the attribute takes each attribute value for the user at the time of the search request or at least one attribute of the search request. The scoring system for search result documents according to claim 10.   When storing the second statistical information in the probability information storage unit, the probability distribution of the second statistical information and the probability distribution of the third statistical information are statistically significant for each document to be searched. And a filter means for prohibiting the storage of the second statistical information in the probability information storage unit when there is no statistically significant difference as a result of the determination. 11. The scoring system for a search result document according to claim 10, wherein the scoring system is a search result document. As a response to the user's current search request, for each document listed in the search results provided by the document search means,
The probability of occurrence of the document selection by the user as the first statistical information respectively generated from the past document selection history of the user, and the current time when selecting the document as the second statistical information Based on the conditional occurrence probability of at least one attribute information that the user or the search request has,
A score calculation device for a search result document, which estimates the degree of probability that matches a user's current search request and quantifies the score as a score.   For each document listed in the search result based on the user's current search request, the occurrence probability of the document selection as the first statistical information and at least one of the second statistical information The score calculation apparatus for a search result document according to claim 13, wherein the score is calculated by integrating and integrating the conditional occurrence probability of the attribute information. As a response to the user's current search request, for each document listed in the search results provided by the document search means,
Probability of occurrence of the document selection by the user as the first statistical information generated from the past document selection history of the user, and the current use when selecting the document as the second statistical information Based on the conditional occurrence probability of at least one attribute information possessed by the person or the search request,
A scoring method for a search result document, comprising a score calculation step for estimating a probability of conformity to a user's current search request and digitizing the score. When selecting the document as the first statistical information as the first statistical information and the document as the second statistical information from the user's past document selection history as the first statistical information Probability information generation step of generating a conditional occurrence probability of each attribute information that the user or the search request has,
As a response to the user's current search request, for each document listed in the search result provided by the document search means, the first statistical information generated in the probability information generation step, respectively, by the user Based on the occurrence probability of document selection and the conditional occurrence probability of at least one attribute information possessed by the current user or the search request when selecting the document as the second statistical information A scoring method for a search result document, comprising: a score calculation step that estimates a probability of conformity to the current search request and quantifies the score as a score.   In the score calculation step, for each document listed in the search result based on the user's current search request, the probability of occurrence of the document selection as the first statistical information and the second statistical information 17. The search result document scoring method according to claim 15, wherein the score is calculated by integrating and integrating the conditional occurrence probability of at least one of the attribute information.   The conditional occurrence probability of the attribute information, as the second statistical information, is calculated for each document to be searched and for at least one attribute of the user or the search request when selecting the document. 17. The search result document scoring method according to claim 15, wherein the attribute is defined as a probability of taking each attribute value. For each document to be searched, the occurrence probability of document selection as the first statistical information and the conditional occurrence probability of each attribute information as the second statistical information respectively generated in the probability information generation step. A probability information storage step of storing in the probability information storage unit;
The scoring of a search result document according to claim 16, wherein, in the score calculation step, the first and second statistical information corresponding to the probability information storage unit is read and the score is calculated. Method. A document selection history storage step of storing a user's past document selection history based on the search result provided from the document search means in a document selection history storage unit;
17. The scoring of a search result document according to claim 16, wherein in the probability information generation step, the document selection history is read from the document selection history storage unit to generate the first and second statistical information. Method. In the probability information generating step,
From the user's past document selection history, for each search target document, the frequency of occurrence of the document selection by the user, and the condition of each attribute information that the user or the search request has when selecting the document After obtaining the frequency of occurrence,
For each document to be searched, the occurrence probability of the document selection as the first statistical information is generated based on the obtained occurrence frequency of the document selection, and the conditional occurrence frequency of each acquired attribute information is set. The search result document scoring method according to claim 16, further comprising: generating a conditional occurrence probability of each attribute information as the second statistical information. In the probability information generating step,
For each document to be searched from the past document selection history of the user, the old occurrence probability of the document selection calculated based on the occurrence frequency of the document selection acquired in the previous past period, and the previous past period Integrating the new occurrence probability of the document selection calculated based on the obtained occurrence frequency of the document selection, generating and updating the occurrence probability of the document selection as the first statistical information, and
For each document to be searched, the old conditional occurrence probability of each attribute information calculated based on the conditional occurrence frequency of each attribute information acquired in the previous past period, and each attribute acquired in the previous past period Integrating the new conditional occurrence probability of each attribute information calculated based on the conditional occurrence frequency of information to generate and update the conditional occurrence probability of each attribute information as the second statistical information; 17. The search result document scoring method according to claim 16, In the probability information generating step,
When generating and updating the occurrence probability of the document selection and the conditional occurrence probability of each attribute information, at least weight the old occurrence probability of the document selection and the old conditional occurrence probability of each attribute information ,
The weight of the old occurrence probability of the document selection or the old conditional occurrence probability of each attribute information is set to be smaller as the value of the old occurrence probability of the document selection or the old conditional occurrence probability of each attribute information is larger. The scoring method for a search result document according to claim 22. In the probability information generating step, the document selection history is read from the document selection history storage unit, and in addition to the first and second statistical information, a user at the time of a search request as third statistical information Or, generate the probability of occurrence of attribute information that the search request has,
In the probability information storage step, the third statistical information generated in the probability information generation step is stored, and
In the score calculation step, when calculating the score for a document, the distribution of conditional occurrence probability of attribute information as the second statistical information and generation of attribute information as the third statistical information If there is no statistically significant difference in the probability distribution,
The score is calculated based on an occurrence probability of the document selection and the third statistical information as the first statistical information instead of the second statistical information. 16. A scoring method for search result documents according to 16.   The probability of occurrence of the attribute information as the third statistical information is defined as the probability that the attribute takes each attribute value for the user at the time of the search request or at least one attribute of the search request. 25. The search result document scoring method according to claim 24.   When storing the second statistical information in the probability information storage unit, the probability distribution of the second statistical information and the probability distribution of the third statistical information are statistically determined for each document to be searched. Determining whether there is a significant difference, and if there is no statistically significant difference as a result of the determination, a filtering step for prohibiting the storage of the second statistical information in the probability information storage unit, 26. The search result document scoring method according to claim 25, further comprising: On the computer,
As a response to the user's current search request, for each document listed in the search results provided by the document search means, the first statistical information generated from the user's past document selection history is used as the first statistical information. Based on the occurrence probability of the document selection by the user and the conditional occurrence probability of at least one attribute information possessed by the current user or the search request when selecting the document as the second statistical information A scoring program for a search result document, which executes a score calculation process for estimating the degree of probability that matches a user's current search request and digitizing it as a score. On the computer,
When selecting the document as the first statistical information as the first statistical information and the document as the second statistical information from the user's past document selection history as the first statistical information Probability information generation processing that generates a conditional occurrence probability of each attribute information that the user or the search request has,
As a response to the user's current search request, by the user as the first statistical information generated by the probability information generation process for each document listed in the search results provided by the document search means Use based on the occurrence probability of the document selection and the conditional occurrence probability of at least one attribute information possessed by the current user or the search request when selecting the document as the second statistical information A scoring program for a search result document, which executes a score calculation process that estimates a degree of probability that matches a user's current search request and digitizes the score.   In the score calculation process, for each document listed in the search result based on the user's current search request, the document selection occurrence probability as the first statistical information and the second 29. The search result document scoring program according to claim 27 or 28, wherein the score is calculated by integrating and integrating the conditional occurrence probability of at least one of the attribute information as statistical information.   The conditional occurrence probability of the attribute information, as the second statistical information, is calculated for each document to be searched and for at least one attribute of the user or the search request when selecting the document. 29. The search result document scoring program according to claim 27 or 28, wherein the attribute is defined as a probability of taking each attribute value. On the computer,
The occurrence probability of document selection as the first statistical information and the conditional occurrence probability of each attribute information as the second statistical information, which are generated by the probability information generation process for each document to be searched The probability information storage process for storing the information in the probability information storage unit, and
29. The search result document according to claim 28, wherein in the score calculation process, the score is calculated by reading out the corresponding first and second statistical information from the probability information storage unit. Scoring program. On the computer,
A document selection history storage process for storing a user's past document selection history based on the search result provided from the document search means in a document selection history storage unit;
29. The search result document according to claim 28, wherein in the probability information generation process, the document selection history is read from the document selection history storage unit to generate the first and second statistical information. Scoring program. In the probability information generation process, the computer
From the user's past document selection history, for each search target document, the frequency of occurrence of the document selection by the user, and the condition of each attribute information that the user or the search request has when selecting the document After getting the occurrence frequency,
For each document to be searched, based on the acquired occurrence frequency of the document selection, the occurrence probability of the document selection as the first statistical information is generated, and based on the conditional occurrence frequency of each acquired attribute information 29. The search result document scoring program according to claim 28, wherein a conditional occurrence probability of each attribute information as the second statistical information is generated. In the probability information generation process, the computer
For each document to be searched from the past document selection history of the user, the old occurrence probability of the document selection calculated based on the occurrence frequency of the document selection acquired in the previous past period, and the previous past period Integrating the new occurrence probability of the document selection calculated based on the frequency of occurrence of the document selection acquired in the step, generating and updating the occurrence probability of the document selection as the first statistical information, and
For each search target document, the old conditional occurrence probability of each attribute information calculated based on the conditional occurrence frequency of each attribute information acquired in the previous past period, and each acquired in the previous past period The conditional occurrence probability of each attribute information as the second statistical information is generated and updated by integrating the new conditional occurrence probability of each attribute information calculated based on the conditional occurrence frequency of the attribute information 30. The search result document scoring program according to claim 28. In the probability information generation process, the computer
When generating and updating the occurrence probability of the document selection and the conditional occurrence probability of each attribute information, weighting weights at least the old occurrence probability of the document selection and the old conditional occurrence probability of each attribute information As the process is executed,
In the weighting process, the larger the value of the old occurrence probability of the document selection or the old conditional occurrence probability of each attribute information is, the smaller the weight of the old occurrence probability of the document selection or the old conditional occurrence probability of each attribute information is. 35. The search result document scoring program according to claim 34, wherein the set processing is executed. In the probability information generation process, the computer
The document selection history is read from the document selection history storage unit, and in addition to the first and second statistical information, a user at the time of the search request or the search request as the third statistical information Generate the occurrence probability of the attribute information you have,
In the probability information storage process, the third statistical information generated in the probability information generation process is stored, and
In the score calculation process, when calculating the score for a document, the distribution of conditional occurrence probability of attribute information as the second statistical information and generation of attribute information as the third statistical information If there is no statistically significant difference in the probability distribution,
The score is calculated based on the occurrence probability of the document selection and the third statistical information as the first statistical information instead of the second statistical information. 28. A scoring program for search result documents according to 28.   The probability of occurrence of the attribute information as the third statistical information is defined as the probability that the attribute takes each attribute value for the user at the time of the search request or at least one attribute of the search request. 37. The search result document scoring program according to claim 36.   When storing the second statistical information in the probability information storage unit in the computer, for each search target document, the probability distribution of the second statistical information and the probability distribution of the third statistical information, A filtering process for determining whether or not there is a statistically significant difference, and forbidding storage of the second statistical information in the probability information storage unit when there is no statistically significant difference as a result of the determination 37. The search result document scoring program according to claim 36, further comprising:

Images