JP2006268789A - Apparatus and method for document retrieval - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To present a result of retrieval, with high quality by reflecting retriever's intention of retrieval. <P>SOLUTION: A history pattern conversion section 105 converts a retrieval intention inputted from a retrieval intention input section 102 by a retriever into a history pattern. A retrieval section 103 calculates a retrieval score of each document based on a retrieval query inputted from a retrieval query input section 101 and further corrects the retrieval score based on the history pattern. A ranking section 104 presents, to the retriever, a result of retrieval based on the corrected retrieval score. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to an electronic search technique.

  There is a technique for storing a history of search contents and operation contents to be used later when searching for electronically stored documents. For example, Patent Document 1 proposes to extract compound word example data from a search history to generate a compound word database. Further, Patent Document 2 proposes to hold search keywords and reference documents together with the date and time, and to record keywords and reference documents that are frequently used in a certain time zone.

  There are many technologies that use the history, but in the past only the date and time were treated as numerical values.

  By the way, when a user refers to a document, the user has an intention. Therefore, it is expected that the quality of search results will be improved if such an intention is used. The present inventor has come to the knowledge that there are many search intentions related to the search history, and has made the present invention. That is, when there is an intention to acquire new information, a history pattern in which a searcher has not searched or referred to a document at least in a range that can be remembered corresponds. When there is an intention to remember predetermined information, a history pattern that a searcher has searched or referred to a document in the past corresponds. In addition to your own history pattern, other people's history patterns may be relevant to your search intent. For example, the intention to acquire information that others know but does not know corresponds to a history pattern that others have searched or referred to in the past but have not searched or referenced in the past.

The present invention not only treats the date and time of the reference history as numeric data, but also associates the history pattern with the search intention to achieve a search in consideration of the intention.
JP 2002-251402 A JP 2003-173352 A

  The present invention has been made in consideration of the above circumstances, and an object thereof is to provide a technique for performing a search reflecting a search intention.

  According to the present invention, in order to achieve the above-mentioned object, the configuration as described in the claims is adopted. Here, prior to describing the invention in detail, supplementary explanations of the claims will be given.

  That is, according to one aspect of the present invention, in order to achieve the above-described object, the document search apparatus includes: a search condition input unit that inputs a search condition; a search intention input unit that inputs a search intention; Search history data generating means for converting the search intention input by the input means into search history data; a document according to the search conditions input by the search condition input means and the search history data generated by the search history data generating means And a document search means for searching.

  In this configuration, it is possible to designate a search intention and perform high-quality search and high-quality search result presentation.

  The search condition may be a keyword or a keyword string, various document attributes may be designated, or a natural sentence or the document itself may be used as the search condition. Natural sentences are decomposed into keywords, for example, and the keywords are used as direct search conditions. When the document itself is used as a search condition, for example, a word vector of the document can be used.

  In this configuration, the document search means may search for documents according to the search condition and narrow down the documents searched according to the search history data.

  For example, a search is performed based on the presence or absence of a keyword, and the search result is narrowed down to that of desired history data.

  Further, the search means may search for a document with a score according to the search condition and adjust the score according to the search history data. The score can be acquired using, for example, a word vector. The adjustment of the score based on the search history data depends on the search intention. For example, in the case of the intention of searching for information to be remembered, a document searched in the past close to the current time is set to have a higher score.

  Further, the search means may search for a document without a score according to the above-mentioned condition, and generate a score for the searched document according to the search history data. As described above, when the intention is to search for information to be remembered, a new scoring is performed so that the closer the search time is to the present time, the higher the score.

  The search intention includes, but is not limited to, an intention to newly discover unknown information and an intention to recall information that has been seen.

  In addition, the search intention may include an intention to obtain information that only the user does not know in the affiliated organization, and an intention to obtain information that is utilized only by the user in the affiliated organization.

  Search intent is generally useful in larger organizations including the affiliated organization, but in the affiliated organization, it is generally useful in the above-mentioned large organization. May include an intention to obtain information that is unknown to the public, and an intention to obtain information that is useful to the organization to which the user belongs but is not generally known to the large organization.

  The present invention can be realized not only as an apparatus or a system but also as a method. Of course, a part of the invention can be configured as software. Of course, a software product used to cause a computer to execute such software is also included in the technical scope of the present invention.

  The above and other aspects of the invention are set forth in the appended claims and are described in detail below using examples.

  According to the present invention, a search result can be acquired with high quality reflecting the search intention.

  Examples of the present invention will be described below.

  FIG. 1 shows a document search system 100 according to an embodiment of the present invention. In this embodiment, the document search system 100 is realized by installing a program 201 in a computer 200. The following functional blocks are realized by cooperating the hardware resources and software resources of the computer 200. In the figure, the document search system 100 is constructed by one computer 200, but the document search system 100 may be constructed by a client server system.

  In FIG. 1, a document search system 100 includes a search query input unit 101, a search intention input unit 102, a search unit 103, a ranking unit 104, a history pattern conversion unit 105, a document database 107, a reference history database 108, a user information database 109, and the like. It is comprised including.

  The search query input unit 101 receives, for example, one or a plurality of keywords from a user using a predetermined user interface. Natural sentences and documents may be input or specified instead of keywords. The search intention input unit 102 inputs a search intention as shown in FIG. 5, for example, and can typically be specified and input using a pull-down menu. In this example, the search intention is as follows, but is not limited thereto. Here, it is assumed that the searcher belongs to a predetermined team and the team is in an entire organization such as a company.

(1) I want to discover new information that I do not know.
(2) I want to remember the information I have seen.
(3) I want to get information that I don't know.
(4) I want to know information that I am only using.
(5) I want to get information that only my team does not know.
(6) I want to get information that only my team uses.
(7) Although it is generally useful, I would like to know the information that only my team uses.
(8) I would like to obtain information that is generally useful but not known within the team.
(9) I want to obtain information that is useful to the team but is not generally known.

  The history pattern conversion unit 105 converts the search intention input from the search intention input unit 102 into a history pattern. For example, in FIG. 5, the history pattern is shown as a pattern indicating which history frequency is positively important, negatively important, or not considered for “self”, “team”, and “whole”. Of course, other expression methods can be adopted. For example, (1) when a search intention is selected to discover new information that you do not know, “self” is “x”, and the others are irrelevant (“−”). Therefore, the score is reduced when the user searches or references. An example of the score operation will be described later.

  The search unit 103 calculates the score of each document based on the input search query, and corrects the score based on the history pattern from the history pattern conversion unit 105. For the score based on the search query, for example, an inner product of a word vector indicating a keyword or a keyword string and a word vector of each document acquired from the document database 107 may be obtained. The document database 107 holds information for score calculation as shown in FIG. 2 for each document, for example, and uses this document vector. The correction of the score uses a history pattern, user information and organization information in the user information database 109, and a reference history from the reference history database 108.

  For example, in FIG. 5, when “self” is “x”, that is, when the frequency of reference is large and the score is decreased, the correction formula is as follows.

(Formula 1)
S0 × (T−Ta) / Tr (T−Tr <= Ta)
S0 (T-Tr> Ta or Ta does not exist)

  However, S0 is the search score of the document calculated from the search query, T is the current time, Ta is the latest time referred to by itself, and Tr is the reference time. In other words, if it exceeds the reference time since it was referred to, it is assumed that it was forgotten and the original score S0 is used as it is. Otherwise, the closer the latest reference time is to the current time, the closer the score To be smaller.

  In FIG. 5, when “I” is “O”, that is, when the frequency referred to by myself is large, when the score is increased, a predetermined number is added to the initial score or multiplied by a predetermined number. Good.

  In FIG. 5, when another person, that is, another person of the team or organization is “x”, processing such as dividing the score by the frequency is performed so as to reduce the score according to the frequency referred to by the other person.

  In FIG. 5, when another person, that is, another person of the team or the organization is a circle, a predetermined number may be added or multiplied by a predetermined number to the initial score.

  It is determined using the user information and organization information record shown in FIG. 4 whether or not the searcher's team is another person and whether or not the searcher's team is the other person in the entire organization. These user information and organization information are held in the user information database 109.

  FIG. 6 shows the operation of the above-described embodiment as a whole, and the processing is as follows.

[Step S10]: A search query and a search intention are input.
[Step S11]: A search score is calculated for each document with reference to the search query.
[Step S12]: The search intention is converted into a history pattern, and the search score is corrected based on the reference history of each document according to the history pattern.
[Step S13]: Ranking is performed in descending order of corrected search scores and presented to the user.

  In addition, this invention is not limited to the above-mentioned Example, A various change is possible in the range which does not deviate from the meaning. For example, the following formula can be used to correct the search score.

(Formula 2)
S0 * (1-C) + ((T-Ta) / Tr) * C (T-Ta <= Tr)
S0 × (1-C) + C (T−Ta> Tr or Ta does not exist)

  Here, C is a scoring weight (0 <= C <= 1), and the larger C is, the more the correction is made to a score that emphasizes the reference history.

  Further, when using another person's reference history, the following equation can be used.

(Formula 3)
S0 × A / Ar (A <= Ar)
S0 (A> Ar)

  However, A is the reference count of others, Ar is the reference count (0 <Ar), and Ar is set in advance. Like the previous example

(Formula 4)
S0 × (1-C) + (A / Ar) × C (A <= Ar)
S0 × (1-C) × C (A> Ar)
And the weight of the reference history may be cut variable.

  Further, if C = 1 in Expression 2, it is possible to deal with a case where there is originally no search score, that is, a case where the search result is a choice between hit and not hit.

  That is, it is as follows.

(Formula 5)
((T-Ta) / Tr) × C (T-Ta <= Tr)
C (T-Ta> Tr or Ta does not exist)

  Further, when the user refers to the document, it is stored in the reference history database 108 of the combination of the user name and the reference date, but when there is a past reference history, the reference date is overwritten (updated writing). Instead of overwriting, all reference histories may be retained, or the reference count and the last reference date may be retained and used.

  First, as shown in FIG. 3, the reference history uses a flag and percentage indicating how much the document has been read, and specifies the search intention from the viewpoint of the document that has been read, the document that has been called halfway, the document that has not been read at all, etc. May be. In this case, the user may input the flag value or the read amount (percentage) at the time of reference, or may be used when the amount can be automatically acquired from the reference application. In addition, when reading to the end, the reference may be empty because the document is sufficiently grasped.

It is a block diagram which shows the Example of this invention as a whole. It is a figure explaining the example of data for calculating the search score of the document of the above-mentioned Example. It is a figure explaining the example of data of the reference history of the above-mentioned Example. It is a figure explaining the example of data of the user information of the above-mentioned Example. It is a figure explaining the example which converts the search intention of the above-mentioned Example into a history pattern. It is a figure explaining operation | movement of the said Example.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 Document search system 101 Search query input part 102 Search intention input part 103 Search part 104 Ranking part 105 History pattern conversion part 107 Document database 108 Reference history database 109 User information database 200 Computer 201 Program

Claims (9)

Search condition input means for inputting search conditions;
A search intention input means for inputting a search intention;
Search history data generating means for converting the search intention input by the search intention input means into search history data;
A document search apparatus comprising: a document search unit that searches for a document according to the search condition input by the search condition input unit and the search history data generated by the search history data generation unit.   2. The document search apparatus according to claim 1, wherein the document search means searches for a document according to the search condition and narrows down the searched documents according to the search history data.   2. The document search apparatus according to claim 1, wherein the search means searches for a document with a score according to the search condition, and adjusts the score according to the search history data.   The document search apparatus according to claim 1, wherein the search means searches for a document without a score according to the condition, and generates a score of the searched document according to the search history data.   The document search apparatus according to claim 1, wherein the search intention includes an intention of newly discovering unknown information and an intention of recalling information that has been seen.   The search intention includes the intention of obtaining information that only the user does not know in the affiliated organization and the intention of obtaining information utilized only by the user in the affiliated organization. The document search device according to any one of the above.   The above search intention is generally useful in larger organizations including the affiliated organization, but in the affiliated organization, the intent to obtain information utilized only by itself is generally useful in the large organization, but within the organization The document search apparatus according to claim 1, which includes an intention to obtain unknown information and an intention to obtain information that is useful in the organization to which the user belongs but is not generally known in the large organization. . A step for the search condition input means to input the search condition;
A step of inputting a search intention by a search intention input means;
A search history data generating means for converting the search intention input by the search intention input means into search history data;
A document search method characterized in that the document search means includes a step of searching for a document according to the search condition input by the search condition input means and the search history data generated by the search history data generation means. Search condition input means for inputting search conditions;
A search intention input means for inputting a search intention;
Search history data generating means for converting the search intention input by the search intention input means into search history data;
It is executed by a computer to realize a document search means for searching for a document according to the search condition input by the search condition input means and the search history data generated by the search history data generation means. Computer program for document retrieval.