JP2011175525A - Retrieval device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To solve the problem of the ranking order of content decreasing even if the content is created by the same person who has created content that is clicked many times, because the number of clicks is zero or a value close to it immediately after it has been registered. <P>SOLUTION: The retrieval device includes a click history storage means 105 storing a history of the numbers of clicks, a retrieval means 101 calculating a first score while taking the number of clicks into consideration for each content retrieved, and creating primary retrieval results ranked by the first score, a reliability calculation means 102 calculating the value of reliability based on the ranking of the created content in the primary retrieval results for every person who has created the contents in the primary retrieval results, and a re-ranking means 103 calculating a second score while taking into consideration the number of clicks and the value of reliability regarding the person who has created content for every retrieved content, and for output of secondary retrieval results ranked by the second score. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a search device having a function of ranking search results.

  In a search device that searches for content that matches a search condition input by a user, various ranking methods for displaying a list of search results have been proposed or put into practical use. One of them is a method in which content viewed many times by many users is ranked high because it is considered to be popular, important and reliable. Hereinafter, this method is referred to as a click feedback method.

  An example of a search device using a click feedback technique is described in Patent Document 1. The search device described in Patent Literature 1 includes a browsing information storage unit that stores a past browsing history for each content. Specifically, the browsing information storage unit stores the number of browsing times and the degree of burst (rate of increase / decrease from the number of browsing the previous day) and the traffic rank derived from these for each content. When searching for content that matches the search conditions from the user, the search device sorts and outputs the search results according to the traffic rank of the content.

  On the other hand, Patent Document 2 describes a search device that corrects a temporary score of a search result obtained by an arbitrary ranking algorithm according to a distance between a searcher's affiliation and a searcher's affiliation. Yes. Specifically, for example, the provisional score of document A is 0.8, the provisional score of document B is 0.9, and the affiliation of the creator of document A is first business headquarters, first business department, first sales department, first department Suppose that the creator of the document B belongs to the second business headquarters, the first business department, the second sales department, and the first section. At this time, if the affiliation of the creator of the document A is the same as that of the searcher, the weight of the document A is the highest, for example 1.0, and the final score of the document A is 0.8 × 1.0 = 0.8. If the creator of the document B is different from the searcher, the weight of the document B is as low as 0.7, for example, and 0.9 × 0.7 = 0.63 is calculated as the final score of the document B. As a result, the document A that has a lower score than the document B in the provisional score is ranked higher than the document B in the final score. This Patent Document 2 is based on the premise that a document created by a person belonging to an organization closer to the organization to which the searcher belongs is more important for the searcher.

JP 2009-48380 A JP 2009-146044 A

  The click feedback method is used as a ranking algorithm for cross-sectional search of structured data and unstructured data scattered across each server in the company, but consideration for content that has just been registered is insufficient. It is. The reason for this is that the number of clicks of content that has just been registered is 0 or close to it, but this is not because it was not important as a search result. This is because it was not appearing until. For this reason, there is content whose number of clicks will increase significantly in the future among the content whose number of clicks is 0 or a value close thereto. A representative example of such content that may be possible is content newly created by a creator who has already created content with a large number of clicks.

  However, in the search device described in Patent Document 1 described above, content creators are not considered in the calculation of scores, and therefore content by the same creator as the creator who has already created content with a large number of clicks is registered. Therefore, if the number of clicks is 0 or a value close to it, the ranking rank is lowered. Further, in the search device described in Patent Document 2 described above, the tentative score is corrected by comparing the affiliation of the searcher and the affiliation of each content creator, so that it is the same as the creator who has already created content with a large number of clicks. The content created by the creator has a problem that if the content creator's affiliation is different from the affiliation of the searcher, the ranking will become lower.

  The object of the present invention is the above-described problem, that is, even if the content is created by the same creator as the creator who created the content with a large number of clicks, the number of clicks is 0 or a value close to it if it is just registered. Therefore, an object of the present invention is to provide a search device that solves the problem of lowering the ranking order.

  A search device according to an aspect of the present invention includes a click history storage unit that stores a history of the number of clicks of each content, searches for content by inputting a user search request, and relates to the content for each searched content. A search means for calculating a first score in consideration of the number of clicks and creating a primary search result ranked according to the calculated first score; and for each creator of content in the primary search result, A reliability calculation means for calculating a reliability value based on the ranking of the created content in the primary search result; and for each searched content, the number of clicks related to the content and the reliability related to the creator of the content. The second score is calculated in consideration of the value, and ranking is performed based on the calculated second score. And a re-ranking means for outputting a search result.

  Since the present invention has the configuration as described above, even if the content is registered soon and the number of clicks is 0 or a value close to it, the content by the same creator as that content is the current search result. If the ranking is high, it is possible to prevent the ranking from becoming low.

It is a block diagram of a 1st embodiment of the present invention. It is a block diagram of the 2nd Embodiment of this invention. It is a figure which shows the example of a calculation of the click score for every content searched in the 2nd Embodiment of this invention, a temporary score, an author confidence coefficient, a 1st score, a 1st ranking, a final score, and a final ranking.

Next, embodiments of the present invention will be described in detail with reference to the drawings.
[First embodiment]
Referring to FIG. 1, a search device 100 according to the first embodiment of the present invention is generally configured from a server device, searches for content that matches a search condition input by a user, and ranks the search results. And has a function of returning it to the user. The search apparatus 100 includes search means 101, reliability calculation means 102, reranking means 103, history update means 104, and click history storage means 105.

  The click history storage unit 105 has a function of storing a history of the number of clicks for each content. The number of clicks has the same meaning as the number of times viewed.

  The search unit 101 has a function of searching for content by inputting a user search request, a function of calculating a first score for each searched content in consideration of the number of clicks related to the content, and a calculated first score And a function of creating a primary search result ranked by score.

  The content to be searched by the search means 101 or its index is stored in a storage device in the search device 100 or an external storage device accessible from the search device 100. When the search device 100 is a device that performs an enterprise search (Enterprise Search), the content to be searched includes various types of structural data such as data in a database, data in a file server, and business applications, Contains structural data. When the search device 100 is a device that performs an Internet search, the content to be searched includes website pages and the like.

  The reliability calculation unit 102 has a function of calculating a reliability value for each creator of the content in the primary search result created by the search unit 101 based on the ranking of the created content in the primary search result.

  The creator of the content is described in the content or index stored in the storage device in the search device 100 or an external storage device accessible from the search device 100. Further, the click history storage means 105 may store the number of clicks of each content and information about the creator.

  For each content creator, the reliability calculation means 102 determines the trust based on the first score given to the content ranked highest in the primary search result among the content created by the creator. You may calculate the degree. Specifically, for the first score given to the highest ranking content in the primary search result, it was given to the content ranked highest in the primary search result among the content created by the content creator The reliability may be calculated based on the ratio of the first score.

  Further, the reliability calculation means 102 may calculate the reliability of the content creator by a method different from the above. For example, for each content creator, the reliability may be calculated based on the average score of the first scores assigned to the content appearing in the primary search result among the content created by the creator.

  The re-ranking unit 103 calculates a second score for each content searched by the search unit 101 in consideration of the number of clicks related to the content and the reliability value related to the creator of the content, And a function of generating and outputting a secondary search result ranked by the calculated second score.

  The re-ranking unit 103 calculates a click score from the number of clicks related to the content for each content searched by the search unit 101, and the author trust is calculated from the calculated click score and the reliability value related to the creator of the content. An auxiliary value may be calculated and a second score may be calculated considering the calculated author confidence auxiliary value. The calculated work trust auxiliary value may be a value that increases as the reliability value increases and decreases as the click score increases. Further, the re-ranking unit 103 calculates a corrected click score from the calculated author trust assist value and the click score, and calculates a second score from the calculated corrected click score and the provisional score for each searched content. You can do it.

  The history update unit 104 has a function of updating the number of clicks of the content recorded in the click history storage unit 105 when the user browses the content. For example, when the user clicks on a certain content in the search result and requests browsing, the click history update unit 104 transmits the requested content to the user, and transmits the requested content stored in the click history storage unit 105. The number of clicks may be increased by one.

  Next, the operation of this embodiment will be described.

  When a search request specifying a search condition such as a search keyword is input from a user terminal such as a client terminal (not shown), the search unit 101 of the search device 100 searches for content that matches the search condition. Next, the search unit 101 calculates a first score for each searched content in consideration of the number of clicks related to the content, and creates a primary search result ranked by the calculated first score.

  Next, the reliability calculation unit 102 calculates a reliability value for each creator of the content in the primary search result created by the search unit 101 based on the ranking of the created content in the primary search result. The author's reliability is based on the ranking of the content created by the creator of the content searched for this search keyword, rather than targeting the entire content created by the creator in the past. Therefore, the calculated reliability indicates the reliability of the creator in the related field of the search keyword.

  Next, the re-ranking unit 103 calculates a second score for each content searched by the search unit 101 in consideration of the number of clicks related to the content and the reliability value related to the creator of the content. A secondary search result ranked with the calculated second score is created and output. The search unit 101 calculates the score (first score) without considering the reliability, whereas the re-ranking unit 103 calculates the score (second score) in consideration of the reliability. Is different. As a result, the search results ranked in consideration of the number of clicks and the reliability of the creator are displayed on the screen of the terminal of the user who made the search request.

  Thereafter, when the user browses any content in the search result, the history update unit 104 updates the number of clicks for each group to which the content belongs stored in the click history storage unit 105.

  As described above, according to the present embodiment, even if the content has just been registered and the number of clicks is 0 or a value close to it, content by the same creator as that content is ranked high in the current search result. It is possible to prevent the ranking order from being lowered if it is ranked. The reason for this is that the content creators ranked higher in the primary search results that rank the content that matches the search keyword in consideration of the number of clicks. This is because the final score of the content created by the creator is calculated in consideration of not only the number of clicks but also the given reliability.

[Second Embodiment]
Next, a second embodiment of the present invention will be described in detail with reference to the drawings.

  Referring to FIG. 2, the second embodiment of the present invention is a client terminal 210 that transmits a search request including a search keyword, and a search that receives a search request, searches for content, creates a search result, and outputs the search result. The server 200, one or more content storage servers 230 that actually store content, and a network 220 that connects these to be communicable with each other.

  The search server 200 includes a provisional ranking score determination unit 201, a click score calculation processing unit 202, a ranking score calculation processing unit 203, an author confidence coefficient calculation processing unit 204, a client request reception / return processing unit 205, and a click history information storage unit 206. It has. In the relationship with FIG. 1, the click history information storage unit 206 corresponds to the click history storage unit 105, and the client request reception / return processing unit 205, the temporary ranking score determination unit 201, the click score calculation processing unit 202, and the ranking score calculation processing. The unit 203 corresponds to the search unit 101, the author reliability coefficient calculation processing unit 204 corresponds to the reliability calculation unit 102, and the ranking score calculation processing unit 203 corresponds to the reranking unit 103.

[Overview]
While the amount of information scattered throughout the enterprise continues to increase, the person in charge of business and the product development department rarely change frequently within a certain period. As a result, in the search for a certain business or product, the content created by the person in charge or the department in charge of the business is considered to be more reliable and important than the content created by others. From the viewpoint of obtaining the reliability and importance evaluation of the newly created content from the reliability of the content author, this embodiment implements a click feedback method in an in-company search using content creator information.

  Specifically, at the time of search, first, a temporary ranking score of a search result is determined by an existing scoring algorithm and a normal click feedback technique (any scoring algorithm is acceptable). Next, the author's confidence coefficient is defined using the provisional ranking score. This confidence coefficient is calculated from the provisional ranking score of the content created by the author. Since the content at the top of the search results has been highly evaluated by users, the author of this content is recognized as having high reliability in the related field of the search keyword, and the reliability coefficient is increased. When the same author has multiple contents for one search keyword, the calculation is performed using the ranking score of the content having the highest provisional ranking score. Then, the ranking score is recalculated by adding the reliability coefficient to the provisional ranking score of the content created by the same author.

  By implementing the click feedback method in this way, even if the newly created content is clicked on, the ranking score value is assisted by the confidence coefficient of the author of the content. As a result, the reliability and importance of the content can be correctly evaluated without depending on the total number of clicks of the content, and the user can be provided with a ranking ranking appropriate to the search result.

  Next, the operation of this embodiment will be described with reference to FIG. As a premise, it is assumed that the provisional ranking score determination unit 201 can use a general-purpose scoring engine, and information necessary for this is already stored in the search server 200.

  When the client terminal 210 transmits a search request including a search keyword to the client request reception / return processing unit 205, the client request reception / return processing unit 205 receives this search request and transmits information to the provisional ranking score determination unit 201. . The temporary ranking score determination unit 201 searches for content that matches the search keyword, and calculates a temporary ranking score. Subsequently, the searched content is transferred to the click score calculation processing unit 202, and the calculated temporary ranking score is transferred to the ranking score calculation processing unit 203.

  The click score calculation processing unit 202 refers to the click history stored in the click history information storage unit 206 and calculates the click score of the searched content. Then, the calculated click score is passed to the ranking score calculation processing unit 203.

  The ranking score calculation processing unit 203 calculates a ranking score using the calculated temporary ranking score and click score. This calculation is performed using Formula 2 to be described later, but since the author confidence coefficient is not created, the author confidence coefficient is calculated with a default value of zero.

  Subsequently, the author confidence coefficient calculation processing unit 204 obtains the ranking score of the content calculated by the click feedback method not considering the author's reliability from the ranking score calculation processing unit 203, and the author's trust is calculated from the ranking score. Calculate the coefficient. When a plurality of contents are created by the same author, the content is calculated from the ranking score of the content having the highest ranking score among the contents created by the author. The higher the ranking score, the higher the confidence factor for the creator of the content. The author confidence coefficient calculation processing unit 204 returns the calculated author confidence coefficient to the ranking score calculation processing unit 203. In the present embodiment, information on the author (creator) of each content is stored in the click history information storage unit 206.

  Finally, the ranking score calculation processing unit 203 recalculates the ranking score with the updated author confidence coefficient, and returns the search result to the client request reception / return processing unit 205. The client request reception / return processing unit 205 sorts the search results in descending order of ranking score and returns them to the client terminal 210.

  The client terminal 210 that has received the search result displays the search result on the screen. The user can browse the content from the content storage server 230 by selecting the content to be browsed from the search result list. At that time, the content storage server 230 updates the number of clicks of the browsed content in the click history information storage unit 206.

  Next, a more specific operation of this embodiment will be described. The operation of this embodiment includes the following four steps.

Step1.
The client request reception / return processing unit 205 receives a search request from the client terminal 210, and the temporary ranking score determination unit 201 lists all the contents that are search results, and calculates the score of each temporary ranking.

  Since the tentative ranking score determination unit 201 can apply an arbitrary algorithm, the description thereof is omitted. The click score calculation processing unit 202 calculates a click score that reflects the number of clicks of the search result content listed by the temporary ranking score determination unit 201 according to the following calculation formula 1. The numerical range of the click score is 0-1.

Click score ≡ Clicks / max (Clicks on all content) (1)
However, max (number of clicks of all contents) means the maximum value among the number of clicks of each content.

Step2.
The ranking score calculation processing unit 203 calculates the temporary ranking score “temporary score” of the search result content calculated by the temporary ranking score determination unit 201 and the click feedback coefficient calculated by the click score calculation processing unit 202 according to the following calculation formula 2. The first ranking score is calculated using “click score”. At this time, since the author confidence coefficient has not been calculated yet, the default value is defined as “0”. That is, the calculation is performed in the same manner as the normal click feedback method (in the case of a search using the normal click feedback method to which the present invention is not applied, this order is returned to the client terminal 210 as the search result as it is).

Ranking score ≡ provisional score x {click score + author confidence coefficient x (1-click score)} (2)

  In this specification, the author confidence coefficient × (1−click score) is defined as the author confidence auxiliary value.

Step 3.
The ranking score calculation processing unit 203 passes the ranking score of the search result content that is the calculation result to the author confidence coefficient calculation processing unit 204. Then, the author confidence coefficient calculation processing unit 204 calculates the author confidence coefficient by the following calculation formula 3. The higher the ranking score calculated in the case of normal search, the higher the confidence factor is attached to the content author. The numerical value range of the author confidence coefficient is 0-1.

Author confidence coefficient ≡ ranking score / max (ranking score of all contents) (3)
However, max (ranking score of all contents) means the maximum ranking score of each content.

Step4.
After the above processing is completed, the author confidence coefficient calculation processing unit 204 returns the author confidence coefficient to the ranking score calculation processing unit 203. The ranking score calculation processing unit 203 calculates a final ranking score reflecting the author's reliability of the content according to the calculation formula 2 using the calculated author confidence coefficient, the provisional score, and the click score. This final ranking score is returned to the client terminal 210.

  When calculating the ranking score using the author confidence coefficient, the ranking score is assisted by the author's confidence coefficient for content with a small number of clicks. The influence of the auxiliary value on is reduced. For this reason, in the calculation formula 2, it is {author trust coefficient × (1−click score)}.

  The user of the client terminal 210 that has received the search result selects (clicks) the content to be browsed, and refers to the content in the content storage server 230. At this time, the selected click history information (content / author) is transmitted to the search server 200 to update the click history information.

  FIG. 3 shows a calculation example of a click score, a provisional score, an author confidence coefficient, a first score, a first ranking, a final score, and a final ranking for each searched content. Content 2 and content 5 are the same author. Content 5 is newly created content and has no click history. Content 2 is content created in the past and has a lot of reference history. When the first ranking score was calculated (not considering the author's reliability), the content 5 had no click history, so the ranking was low, but the ranking of the content 2 was higher. If the author's reliability can be highly evaluated from the content 2 and the final ranking score is calculated (considering the author's reliability), there is no significant difference between the provisional scores of the two contents 2 and 5 It can be seen that the content 5 is ranked higher than the content 2.

  According to this embodiment, the effects described below can be obtained.

  The normal click feedback method tends to calculate the ranking score of content having no click history, and there is a problem that the ranking ranking of newly created content cannot be evaluated correctly. However, in the present embodiment, on the premise of a normal click feedback method, the reliability of the content creator is reflected in the content ranking score, and the ranking ranking of the content is determined together with the click history. For this reason, even if the content is newly created, if the content by the same creator as that content is ranked high in this search result, the ranking will increase and there will be many opportunities for the searcher to see it. can do.

  As mentioned above, although embodiment of this invention was described, this invention is not limited to the above example, Various other addition changes are possible. In addition, the search device of the present invention can be realized by a computer and a program as well as the functions of the search device according to hardware. The program is provided by being recorded on a computer-readable recording medium such as a magnetic disk or a semiconductor memory, and is read by the computer at the time of starting up the computer, etc. Function as a search device and a search server.

  It can be applied to the use of a ranking algorithm for cross-sectional search of structured data and unstructured data scattered in each server in a company.

DESCRIPTION OF SYMBOLS 100 Search apparatus 101 Search means 102 Reliability calculation means 103 Reranking means 104 Click history update means 104 History update means 105 Click history storage means 120 Search means 200 Search server 201 Temporary ranking score determination part 202 Click score calculation process part 203 Ranking score Calculation processing unit 204 Author confidence coefficient calculation processing unit 205 Return processing unit 206 Click history information storage unit 210 Client terminal 220 Network 230 Content storage server

Claims (8)

Click history storage means for storing a history of the number of clicks of each content;
A search is performed by inputting a user's search request, and a first score is calculated for each searched content, taking into account the number of clicks related to the content, and ranking is performed based on the calculated first score. A search means for creating search results;
For each creator of content in the primary search result, reliability calculation means for calculating a reliability value based on the ranking of the created content in the primary search result;
For each searched content, a second score is calculated in consideration of the number of clicks related to the content and the reliability value related to the creator of the content, and ranking is performed based on the calculated second score. A search device comprising re-ranking means for outputting a secondary search result. The reliability calculation means, for each content creator, based on the first score given to the content ranked highest among the primary search results among the content created by the content creator The search device according to claim 1, wherein the reliability is calculated. In the primary search result among the contents created by the content creator for the first score assigned to the highest ranking content in the primary search result for each content creator. The search device according to claim 1, wherein the reliability is calculated based on a ratio of the first score given to the content ranked highest. The re-ranking means calculates a click score from the number of clicks related to the content for each searched content, and supports author trust from the calculated click score and the reliability value related to the creator of the content. 4. The search device according to claim 1, wherein a value is calculated, and the second score is calculated in consideration of the calculated author confidence auxiliary value. 5. The re-ranking unit calculates a click score from the number of clicks related to the content for each searched content, and the reliability is calculated from the calculated click score and the reliability value related to the creator of the content. An author confidence assist value that increases as the degree value increases and decreases as the click score increases, calculates a corrected click score from the calculated author confidence assist value and the click score, and calculates 5. The search device according to claim 1, wherein the second score is calculated in consideration of a corrected click score. 6. The re-ranking unit calculates a click score from the number of clicks related to the content for each searched content, and the reliability is calculated from the calculated click score and the reliability value related to the creator of the content. An author confidence assist value that increases as the degree value increases and decreases as the click score increases, calculates a corrected click score from the calculated author confidence assist value and the click score, and calculates The search device according to claim 1, wherein the second score is calculated from a corrected click score and a provisional score for each searched content. A search method that is executed by a search device that includes a click history storage unit that stores a history of the number of clicks of each content, a search unit, a reliability calculation unit, and a reranking unit,
The search means inputs a user search request to search for content, calculates a first score in consideration of the number of clicks related to the content for each searched content, and the calculated first score Create the primary search results ranked in
The reliability calculation means calculates a reliability value for each creator of the content in the primary search result based on the ranking of the created content in the primary search result,
The re-ranking means calculates a second score for each searched content in consideration of the number of clicks related to the content and the reliability value related to the creator of the content. A secondary search result ranked by a score of 2 is output. A computer provided with a click history storage means for storing a history of the number of clicks of each content,
A search is performed by inputting a user's search request, and a first score is calculated for each searched content, taking into account the number of clicks related to the content, and ranking is performed based on the calculated first score. A search means for creating a search result; a reliability calculation means for calculating a reliability value based on a ranking of the created content in the primary search result for each creator of the content in the primary search result;
For each searched content, a second score is calculated in consideration of the number of clicks related to the content and the reliability value related to the creator of the content, and ranking is performed based on the calculated second score. A program for functioning as a re-ranking means for outputting secondary search results.

Images