WO2010008047A1

WO2010008047A1 - Integrated evaluation device, integrated evaluation system, integrated evaluation method, integrated evaluation program, and recording medium

Info

Publication number: WO2010008047A1
Application number: PCT/JP2009/062892
Authority: WO
Inventors: 孝白木; 威有熊; 展久白石; 洋一永井
Original assignee: 日本電気株式会社
Priority date: 2008-07-17
Filing date: 2009-07-16
Publication date: 2010-01-21

Abstract

Provided are an integrated evaluation device, an integrated evaluation system, an integrated evaluation method, an integrated evaluation program, and a recording medium which are capable of obtaining the result of the integrated evaluation in which the influence by an evaluation device for performing more unique evaluation is emphasized by giving less weight to the results by evaluation devices with high similarity when allocating importance in the integration of the results of the evaluation. The integrated evaluation device is provided with a means connected to a plurality of evaluation devices for performing evaluation according to evaluation conditions, a means for requesting evaluation under evaluation conditions inputted from user terminals to the evaluation devices, and an integration means for integrating by combining the results of the evaluation obtained from the evaluation devices, characterized in that the integrated evaluation device comprises an evaluation-device-similarity calculating means for calculating the similarity between the plurality of evaluation devices, and a weight setting means for setting the weight of each of the evaluation devices with less weight given to evaluation devices with high similarity in consideration of the calculated similarity, and the integration means integrates the results of the evaluation in consideration of the set weights.

Description

Integrated evaluation apparatus, integrated evaluation system, integrated evaluation method, integrated evaluation program, and recording medium

The present invention relates to an integrated evaluation apparatus, an integrated evaluation system, an integrated evaluation method, an integrated evaluation program, and a recording medium, and in particular, an integrated evaluation apparatus, an integrated evaluation system, an integrated evaluation method, and an integrated evaluation program that take into account the similarity between the evaluation apparatuses. And a recording medium.

A search method for extracting data satisfying a search request from data stored in a search target DB (Date Base) is known. In such a search method, search for multimedia data such as images, videos, music, etc., and text search are performed. In the search, a search may be performed by combining not only one feature quantity but also a plurality of feature quantities. Taking an image search as an example, this corresponds to the case of extracting an image that satisfies the search request for both color and pattern (see Patent Document 1).

In addition, even if the search target is not only a photo, but also a variety of types of documents such as a combination of photos, drawings, text documents, etc., in order to obtain useful search results, photo image search and text search An image search system has been proposed in which the scores obtained from the above are integrated, and the obtained overall score is extracted as a candidate image (see Patent Document 2). Also, a search device that calculates a relationship between a problem and a candidate when searching for a case that satisfies the most demand from the case base has been proposed (see Patent Document 3).

Also, an integrated search called meta search and blend search that combine output results of a plurality of evaluation systems is known (Non-Patent Document 1). Since the range of the evaluation target of each evaluation system may be different, the coverage of the evaluation target can be expanded by an integrated search combining a plurality. In addition, when relying on a certain search system, the evaluation system may censor arbitrarily lower the evaluation of an evaluation target, but it can be prevented by an integrated search.

In integrated search, it is possible to distinguish the degree of influence of each evaluation system according to its accuracy and to set it according to the individual and the situation. Therefore, a calculation method of importance is required. Non-Patent Document 2 describes a point addition method for calculating importance based on a user's operation history. By using this method, points can be added to the search system that has influenced the item selected by the user. Furthermore, it is considered that the result of collecting and optimizing evaluation values by many users depends on the similarity of the evaluation system. It is also known that the similarity between ranking results output by each evaluation system is calculated using Kendall's rank correlation coefficient (Non-Patent Document 3) or the like.

JP 2001-134484 A JP 2007-172077 A Japanese Patent No. 3311778

In the integrated search described above, the following cases can be considered when the point addition method for calculating the importance based on the user operation history is used. For example, when there are three different evaluation systems A, B, and C, when a user's action history likes evaluation system A, and it is determined that B and C are not preferred, the integration is optimized for this user In the search, it can be considered that the importance of the evaluation system A is raised and the importance of B and C is lowered (case 1). However, there are almost the same evaluation systems A ', A' ', A' '' as A, and when integrating six different evaluation systems with A, A ', A' ', A' '', B, C. When it is determined that a user's action history likes evaluation system A and does not like B or C, not only A but also A ', A' ', and A' '', which are equivalent to A, are increased in importance. , B and C will be reduced (Case 2). In

cases

1 and 2, evaluation systems A, B, and C are relative to each other depending on the presence / absence of A ′, A ″, and A ′ ″, although the contents of individual systems remain unchanged. It is inevitable to change. There are the following problems in this situation.

The first problem was that there was no way to adjust the distribution in consideration of the algorithm relationship between the evaluation systems. The problem is particularly serious when there are a large number of evaluation systems that can be used so much as spam.

The second problem is that there is no way to adjust the distribution of multiple evaluation systems in consideration of the similarity among the relationships of the evaluation systems. There was a tendency to raise and lower, and it had a sensitive effect on the overall result, and the integrated system could fall into a situation where it could not demonstrate the benefits of ensuring diversity. The situation where this problem occurs is almost the same as the first problem. For example, an integrated system generally recruits evaluation systems, and Company A uses Evaluation System A and Company B uses Evaluation System B. When Company C applies for evaluation system C, Company A is unfair by applying for A ', A' ', A' '' with the same content as A in order to increase their influence. There is a possibility of spam rating system to try to increase its influence on.

The third problem is that the distribution of a plurality of evaluation systems may be adjusted by reflecting the history of selecting user information, but accurate adjustment is difficult without the amount of user history, and the user When collecting a large number of histories, there is a scalability problem, and it was not possible to optimize effectively at the stage where the user's history is not acquired or at the stage where the history is small.

The fourth problem is that the evaluation system spam is effective in the same way as the second problem because the similarity of the evaluation system is not considered even when the user's history called relevance feedback is dealt with immediately. Is in an environment where

The fifth problem is that even when the user's history called relevance feedback is dealt with immediately, the similarity of the search device is not taken into consideration, so that the integrated evaluation system is the same as the third problem. Convergence to optimize according to was slow.

The present invention has been made in view of such a situation, and when the results of a plurality of evaluation devices are integrated to obtain a result, the similarity between the evaluation devices is taken into consideration and the importance is distributed at the time of integration. The purpose of this method is to obtain an integrated evaluation result in which the influence of the evaluation device performing a more original evaluation is deepened by suppressing the weight of the result by the evaluation device having a high degree of similarity.

An integrated evaluation apparatus according to the present invention includes: means for connecting to a plurality of evaluation apparatuses that perform evaluation according to evaluation conditions; means for requesting the evaluation apparatus for evaluation under evaluation conditions input from a user terminal; and An integrated evaluation device comprising an integration means for combining and integrating the acquired evaluation results, considering the similarity between the evaluation device similarity calculation means for calculating the similarity between the plurality of evaluation devices, Weight setting means for setting the weight of each evaluation apparatus by pressing the weight of the evaluation apparatus having a high degree of similarity, and the integration means integrates the evaluation results in consideration of the set weight. And

A first integrated evaluation system according to the present invention includes the integrated evaluation apparatus according to the present invention, a plurality of evaluation apparatuses, and a user terminal.

A second integrated evaluation system according to the present invention is the integrated evaluation apparatus according to the present invention, and further acquires user behavior information for the integrated evaluation result created by integrating the evaluation results by the integrating means. And a user action history storage means for storing the acquired user action information, wherein the integration means stores the user action history storage means for the user who has made an evaluation request under the evaluation condition. If it is stored in, further considering the stored user behavior information history, comprising an integrated evaluation device characterized by integrating the evaluation results, a plurality of evaluation devices, and a user terminal, The user terminal includes user behavior input means for sending information of user behavior performed on the integrated evaluation result acquired from the integrated evaluation device to the integrated evaluation device. Characterized in that it obtain.

The integrated evaluation method according to the present invention includes a step of requesting a plurality of evaluation devices for evaluation under an evaluation condition input from a user terminal, and a step of obtaining an evaluation result corresponding to the request from the plurality of evaluation devices; The step of calculating the similarity between the plurality of evaluation devices, the step of setting the weight of each evaluation device by pressing the weight of the evaluation device having a high degree of similarity in consideration of the calculated similarity, and the set And integrating the evaluation results in consideration of the weights.

The integrated evaluation program according to the present invention includes a process for requesting a plurality of evaluation apparatuses to perform an evaluation under an evaluation condition input from a user terminal, and a process for obtaining an evaluation result corresponding to the request from the plurality of evaluation apparatuses. A process for calculating the similarity between the plurality of evaluation apparatuses, a process for setting the weight of each evaluation apparatus by pressing the weight of the evaluation apparatus having a high degree of similarity in consideration of the calculated similarity. And a process of integrating the evaluation results in consideration of the weights.

The recording medium according to the present invention is a computer-readable recording medium that records the processing of the integrated evaluation program of the present invention.

According to the present invention, when the results of a plurality of evaluation devices are integrated to obtain a result, the similarity between the evaluation devices is taken into consideration, and the result of the evaluation device having a high similarity is used for the importance distribution at the time of integration. By suppressing the weight of, it is possible to obtain an integrated evaluation result in which the influence of the evaluation device that performs more independent evaluation is deepened.

1 is a schematic configuration diagram of an integrated search system according to an embodiment of the present invention. It is a flowchart which shows the operation | movement process of the integrated search apparatus which concerns on embodiment of this invention. It is a figure which shows the ordering of the information which each search engine which concerns on embodiment of this invention output. It is a figure which shows the similarity between each search engine calculated from the ordering of FIG. It is the figure which converted the ordering of FIG. 3 into the score. It is a figure which shows the output of the integrated search result which concerns on embodiment of this invention. It is a schematic block diagram of the integrated search system which concerns on other embodiment of this invention. It is a schematic block diagram of the integrated search system which concerns on other embodiment of this invention. It is a figure which shows the output of the integrated search result which concerns on other embodiment of this invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. The embodiments described below are preferred embodiments of the present invention, and thus various technically preferable limitations are given. However, the scope of the present invention is particularly limited in the following description. As long as there is no description of the effect, it is not restricted to these aspects.

In this specification, the plurality of evaluation devices may be evaluation devices that perform evaluation with a broad meaning such as importance (score) and ordering of information according to evaluation conditions, such as a search engine and an inference engine. The integrated evaluation device is a device that integrates outputs from the plurality of evaluation devices. The system is not limited to an integrated device such as an evaluation device or an integrated evaluation device, and may be a system (evaluation system, integrated evaluation system) composed of a plurality of devices having these functions as a whole. In the following description, a plurality of search engines that output search results according to search conditions are connected, and an integrated search device that integrates such search results will be described as an example of an integrated evaluation device. It is not limited to.

The integrated search system according to the present embodiment includes at least a terminal, an integrated search device, and a search engine. The integrated search device considers the similarity of each search engine when integrating search results from each search engine. By setting the weight of each search engine, it is possible to reduce the importance of search engines that have many similarities (search engines with high similarity) and to integrate the influence of search engines that perform more unique searches. The search result can be obtained.

The calculation of the relative importance between the search engines can also be used in the compatibility feedback system reflecting the result of the information selection of the user who receives the service, and the search engine is selected by the user in the past. Even if the importance of is increased, it can be considered that an appropriate integrated search result is obtained by suppressing the increase in importance of a search engine having many similar ones.

Hereinafter, the integrated search system according to the embodiment of the present invention will be described in detail.

[First Embodiment]
FIG. 1 is a schematic configuration diagram of an integrated search system according to the present embodiment. The integrated search system includes a terminal 1, an integrated search device 2, and a plurality of search engines A to D (reference numerals 3 to 6). In the present embodiment, the terminal 1 and the integrated search device 2, and the integrated search device 2 and each search engine are connected via a network as an example, but the present invention is not limited to this, What is necessary is just to be able to transmit and receive various information regardless of wireless.

In the present embodiment, four search engines are used (search engines A to D), but this is an example, and two or more different search engines may be used. Here, “different” means that the search method and the search target DB are different, for example, each is calculated by a different algorithm.

The terminal 1 includes a search request input unit 11 and a search result output unit 12. Although this embodiment has been described using one terminal, the present invention is not limited to this.

The search request input unit 11 included in the terminal 1 transmits the input search conditions to the search request reception unit 21 included in the integrated search device 2 via the network 9.

Here, the search conditions input to the search request unit 11 include search information, profile information such as the user's gender and age, presence information such as the location information of the terminal 1 possessed by the user, and preferences representing user preferences. Information may be used. When the position information is used as a search condition, for example, the current position is acquired from GPS (Global Positioning Systems) of the terminal 1, and a search for the position is requested. The preference information includes various information such as information acquired by the terminal 1 from past user operations, such as search requests made in the past, Internet browsing history, and other user preference information registered in advance, and is not particularly limited. Also, a combination of search conditions such as preference information and search keywords may be used. For example, when requesting the search keyword “Shinjuku”, if the preference information is also transmitted to the search request receiving unit 21, each search engine is data similar to “Shinjuku”, and the search result further considers the preference information. Is output. For example, if the preference information is information such as “I like exercise”, “I like sweet things” or “I do n’t like classics”, it is information about sports facilities and sweetness information that satisfy the search condition of the keyword “Shinjuku”. Outputs event information etc.

The search result output unit 12 provided in the terminal 1 receives the search results in the integrated search device 2 from the search result integration unit 27 provided in the integrated search device 2, integrates these, and outputs them.

The integrated search device 2 is a device that integrates the output results of the search engine A (3), the search engine B (4), the search engine C (5), and the search engine D (6) and outputs them to the terminal 1. The integrated search device 2 includes a search request reception unit 21, a search request unit 22, a search result reception unit 23, a search engine similarity calculation unit 24, a user behavior acquisition unit 25, a weight calculation unit 26, a search engine similarity DB 261, and a search result. The integration unit 27 is configured.

Next, the operation processing of the integrated search device according to the present embodiment will be described with reference to FIG. FIG. 2 is a flowchart showing an operation process of the federated search apparatus according to the present embodiment.

The search request receiving unit 21 transmits the search conditions received from the search request input unit 11 included in the terminal 1 to the search request unit 22 (steps S1 and S2).

The search request unit 22 transmits the search request contents (search conditions) received from the search request receiving unit 21 to the search engines A to D to be connected via the network 10 (step S3).

The search result reception unit 23 transmits the search results received from the search engines A to D to the search engine similarity calculation unit 24 and the search result integration unit 27 (step S4). Details of the search results in the search engines A to D will be described later.

The search engine similarity calculation unit 24 calculates the similarity between search engines based on the search results of the search engines A to D. Details of similarity calculation between search engines will be described later.

The calculation result (similarity data between search engines) in the search engine similarity calculation unit 24 is input to the search engine similarity DB 261. The search engine similarity DB 261 stores similarity data between search engines received from the search engine similarity calculation unit 24. When the search engine similarity calculation unit 24 inputs the “similarity data between search engines” to the search engine similarity DB 261, the search engine similarity calculation unit 24 confirms that there is update data in the search engine similarity DB 261 with respect to the weight calculation unit 26. Notification is made (step S5).

The weight calculation unit 26 notified of the presence of the update data inquires the search engine similarity DB 261 and acquires the similarity data between the search engines (step S6). The weight calculation unit 26 calculates the weight of each search engine in consideration of the acquired similarity data between the search engines, and transmits the weight to the search result integration unit 27 (step S7). Details of weight calculation of each search engine will be described later.

The search result integration unit 27 integrates the search result of each search engine acquired from the search result reception unit 23 and the weight of each search engine acquired from the weight calculation unit 26, and calculates an integrated search result (step S8). . Details of the integrated search result calculation will be described later.

The calculated integrated search result is transmitted to the search result output unit 12 provided in the terminal 1 (step S9). The search result output unit 12 of the terminal 1 outputs the integrated search result sent from the integrated search device 2 and provides it to the user.

Next, each process will be described in detail. In the present embodiment, the search condition will be described assuming that, for example, a search keyword “mobile phone” is input.

[Search Results in Search Engines A to D (FIG. 2, Steps S3, 4)]
The search engines A to D search for information to be searched for the search request by the search keyword “mobile phone” received from the search request unit 22 according to the respective algorithms, and attach the scores according to the scores or the scores. The received order is transmitted to the search result receiving unit 23. In the present embodiment, the same five filtered information (information a to e) is output in different orders as the result of the search engine A to D outputting information similar to the search keyword “mobile phone” (search result). And An example of each search result ordered by the search engines A to D is shown in FIG. The ordering shown in FIG. 3 is information that is highly evaluated by each search engine as the value is smaller.

[Similarity calculation between search engines (FIG. 2, step S5)]
The search engine similarity calculation unit 24 calculates the similarity for each pair (A and B, A and C, A and D, etc.) of all search engines. Various methods for calculating the degree of similarity can be considered. Here, for example, the degree of correlation is represented by a real number between 0 and 1 using Kendall's rank correlation coefficient (see Non-Patent Document 2). 0 is uncorrelated (reverse), 1 is exactly the same.

FIG. 4 shows the results (degree of correlation) for the permutations of the search results of each search engine shown in FIG. For example, as shown in FIG. 3, the search result of the search engine A is evaluated as “information a> b> c> d> e”, whereas the search result of the search engine B is “information a> b> c”. > E> d ”. In this case, using Kendall's rank correlation coefficient, the similarity between the search engines A and B is “0.9” (see FIG. 4). The similarity data calculated by the search engine similarity calculation unit 24 is input to the search engine similarity DB 261.

The similarity calculation method is not limited to Kendall's rank correlation coefficient, and an inner product of vectors of each permutation may be used. Moreover, the real number from 0 to 1 is an example, and the present invention is not limited to this.

[Calculation of Weight (FIG. 2, Step S7)]
The weight calculation unit 26 sets the vector s as s = (1/4, 1/4, 1/4, 1/4) ^T, and obtains s-cMs. M is a matrix of similarity between the search engines shown in FIG. 4, and c is a constant. If c = 0.2, it can be calculated as in the following formula 1.

When normalization is performed so that the sum of the elements calculated in the above equation 1 becomes 1, the vector w becomes the following equation 2.

The vector w shown in Equation 2 is the weight of each search engine. That is, in the vector s, the weight is uniform and tied (for example, all search engines A to D are 0.25), but the weights corrected in consideration of the similarity of each search engine (search engine A; 0.243, B; 0.229, C; 0.243, D; 0.285). The weight calculation unit 26 transmits the calculated weight data of each search engine to the search result integration unit 27.

[Calculation of Integrated Search Result (FIG. 2, Step S8)]
FIG. 5 shows a matrix M ′ obtained by converting the permutation of information a to e in each search result shown in FIG. 3 into points. The search result integration unit 27 calculates M′w and obtains the integration of search results considering the weight of each search engine.

According to the above formula 3, information b, information a, information c, information e, and information d can be ordered in descending order. The search result integration unit 27 transmits the integrated search result to the search result output unit 12.

The search result output unit 12 outputs the integrated search result sent from the integrated search device 2 as shown in FIG. 6, for example, and presents it to the user.

By integrating the search results from the respective search engines according to the present embodiment, the importance of the search engines having many similarities (highly similar search engines) is increased by considering the weights of the respective search engines. You can obtain integrated search results that reduce the influence of the search engine (reduce the weight) and intensify the influence of search engines that perform more unique searches. This also has the effect of reducing adverse effects on spam-like registrations that only increase the influence of search engines.

[Second Embodiment]
As another embodiment of the present invention, in the similarity calculation between the search engines in step S5 in FIG. 2, not only the search result for the currently executed search condition but also at least one or more search results executed in the past are referred to. Then, the similarity is calculated.

FIG. 7 shows a schematic configuration of the integrated search system according to the present embodiment. The difference from FIG. 1 is that the integrated search device 2 includes a search history DB 262. The search history DB 262 stores search results from each search engine for each search condition requested in the past. Note that the search result history may be stored in various units such as for each user context, for each user, for each user group, or all of them. In consideration of the storage capacity of the search history DB, the search history DB may be stored for a certain period, or a certain amount or more may be transferred to another storage device for storage.

When calculating the similarity between the search engines, the search engine similarity calculation unit 24 refers to each search result in at least one or more search conditions executed in the past, in addition to the search results for the currently executed search conditions. Then, the similarity is calculated.

This can further reduce the importance of search engines with many similar search engines, and obtain integrated search results with a deeper influence from search engines that perform more unique searches.

When a plurality of search histories are extracted from the search history DB 262, the search history filtered by the preference information and context of the user who uses the terminal 1 that has made the search request is extracted, and the similarity of the search engines is calculated. Thus, it is also possible to set the weight of the search engine in consideration of the similarity when matched to the user or the context. If the user preference information is acquired when a search request is received from the terminal 1, the latest preference information can be referred to.

[Third Embodiment]
In the second embodiment, every time a search request is made and a search result is received, the search history is acquired from the search history DB and the similarity between the search engines is calculated. In the configuration shown, the search engine similarity DB 261 may store a history of similarity data input in the past and refer to it when calculating weights. The history of similarity data may be stored in various units such as for each user context, for each user, for each user group, or all of them. Further, in consideration of the storage capacity of the search engine similarity DB 261, the search engine similarity DB 261 may be stored for a certain period, or a certain amount or more may be stored in another storage device. Alternatively, the search history filtered by the preference information or context of the user who uses the terminal 1 that has requested the search when extracting the history of similarity data may be extracted. This makes it possible to obtain an integrated search result in which the influence of a search engine that performs a unique search is deepened.

[Fourth Embodiment]
It is also possible to calculate the similarity between the search engines based on the ordering of all information that is the search target of each search engine without narrowing down by the search condition. It is also possible to calculate the similarity between the search engines based on the search target range of each search engine.

As described above, as shown in the second to fourth embodiments, there are various ways of calculating the similarity between the search devices.

[Fifth Embodiment]
Next, another embodiment of the present invention will be described. This embodiment is an integrated search device that uses relevance feedback, that is, a user operation (such as clicking on a search result list) performed on a search result output to the search result output unit 12 of the terminal 1 or purchasing activity. In an integrated search device that feeds back user behavior information to the weight setting, the weight setting that considers the similarity between the search engines is further integrated to easily increase the influence of the search engine that performs the original search. A search result is obtained.

The feedback to the weight setting of the user behavior information performed on the search result is obtained by, for example, each search engine that acquires and stores the click behavior when the user clicks on the search result displayed on the terminal 1 Relevance feedback is performed by reflecting it in the weight of. This makes the weight setting closer to the user's preference, but in order to obtain integrated search results that tend to deepen the influence of the search engine that performs the original search, suppress how to increase the importance (weight). This is also necessary.

FIG. 8 shows a schematic configuration of the integrated search system according to the present embodiment. As shown in FIG. 8, the present embodiment is particularly characterized in that the terminal 1 includes a user behavior input unit 13, and the integrated search device 2 includes a user behavior acquisition unit 25 and a user behavior history DB 263.

The user behavior input unit 13 included in the terminal 1 is configured to retrieve user behavior information when the user's operation performed on the search result output unit 12 or a user behavior such as a user purchase activity is performed. Is transmitted to the user action acquisition unit 25. Note that the user behavior information is transmitted from the terminal 1 in addition to the case where the user behavior is performed. The user behavior information is temporarily stored in the terminal 1, and the integrated search device 2 is used every predetermined period or at the time of a search request. May be communicated to.

The user behavior acquisition unit 25 included in the integrated search device 2 inputs the user behavior information received from the user behavior input unit 13 to the user behavior history DB 263. At the time of input, the fact that there is update data is transmitted to the weight calculation unit 26. The storage in the user behavior history DB 263 may be provided with an identification number for each terminal, and user behavior information may be managed for each same identification number.

The user behavior history DB 262 provided in the integrated search device 2 stores the user behavior information received from the user behavior acquisition unit 25.

The operation process of this embodiment is as follows. The general flow from acceptance of a search request to calculation of an integrated search result is the same as the processing shown in FIG. 2, but this embodiment is particularly characterized by the processing of the weight calculation unit 26 in steps S6 and S7 of FIG. When calculating the weight of each search engine, the weight calculation unit 26 acquires user behavior information from the user behavior history DB when notified by the user behavior acquisition unit 25 that there is update data. The weight calculator 26 calculates the weight of each search engine based on the similarity data between the search engines and the user behavior information.

The calculation of the weight will be further described in detail. This embodiment will be described using the values in FIGS. 3 to 6 as an example. For example, if the user behavior information is information that the information b ranked at the top in FIG. 6 is clicked, feedback is performed to increase the weight of the search engine that has affected the information b. However, the weight is calculated in consideration of similar data between search engines. Here, the calculation is performed using the weight w of each search engine that has already been calculated in consideration of similar data between search engines (see Equation 2 above). That is, when the score vector sb = (0.8, 0.8, 0.4, 0.8) ^T of each search engine for the information b (see FIG. 6), the value of w shown in the above equation 2 and c ′ ₁ = 0.1, c ′ ₂ Assuming = 0.06, w + c ′ ₁ s−c ′ ₂ M ′s is calculated.

The solution obtained by Equation 4 is normalized so that the sum of vector elements becomes 1, as shown in Equation 5 below, and set as the weight of each search engine.

This result is the same as the weight vector w of each search engine in advance (see Equation 2 above). However, in the present embodiment, D is more unique than A and B, which have search engines with a high degree of similarity, in consideration of the similarity.

FIG. 9 shows the calculation result (integrated search result) of the total score reflecting the weight shown in Equation 5 above. Comparing FIG. 6 with the new FIG. 9, in search engines A, B, and D where the score of information b is the same at 0.8, there are relatively few D similar search engines (see FIG. 4). Information e has moved to the top because it gained a large weight.

In FIG. 9, when information other than the information b is selected (click, purchase, etc.), the same processing can be repeated, and weight distribution can be made in consideration of user behavior and similarity.

Note that it may be reflected by executing a sum vector obtained by adding score vectors s of a plurality of user behavior information at the same time, and is not particularly limited.

According to the above embodiment, even if a large number of user behavior histories are collected in the integrated search device using relevance feedback, there is a possibility that the calculation amount is not scalable, but this can be avoided by the present invention. In addition, weight blending considering similarity can be realized even in a stage where many action histories that indirectly describe preferences and intentions such as information selection by the user are not collected.

The program for the CPU to execute the processing shown in the flowchart of FIG. 2 and the processing in each embodiment constitutes a program according to the present invention. As a recording medium for recording the program, a semiconductor storage unit, an optical and / or magnetic storage unit, or the like can be used. By using such a program and a recording medium in a system having a configuration different from that of each of the above-described embodiments and causing the CPU to execute the program, substantially the same effect as the present invention can be obtained.

As described above, the present invention has been specifically described based on the preferred embodiments, but the present invention is not limited to the above-described ones, and it is needless to say that various modifications can be made without departing from the scope of the present invention.

This application claims priority based on Japanese Patent Application No. 2008-188483 filed on July 17, 2008, the entire disclosure of which is incorporated herein.

1 Terminal 2 Integrated Search Device 3 Search Engine A
4 Search Engine B
5 Search engine C
6 Search Engine D
9, 10 Network 11 Search request input unit 12 Search result output unit 13 User behavior input unit 21 Search request reception unit 22 Search request unit 23 Search result reception unit 24 Search engine similarity calculation unit 25 User behavior acquisition unit 26 Weight calculation unit 27 Search result integration unit 261 Search engine similarity DB
262 Search history DB
263 User behavior history DB

Claims

The means for connecting to a plurality of evaluation devices that perform evaluation according to the evaluation conditions, the means for requesting the evaluation device for evaluation under the evaluation conditions input from the user terminal, and the evaluation results acquired from the evaluation device are combined and integrated. An integrated evaluation device comprising an integration means,
Evaluation device similarity calculation means for calculating the similarity between the plurality of evaluation devices;
Considering the calculated similarity, weight setting means for setting the weight of each evaluation device by pressing the weight of the evaluation device having a high similarity,
The integrated evaluation apparatus, wherein the integration unit integrates the evaluation results in consideration of the set weight.
2. The integrated evaluation apparatus according to claim 1, wherein the evaluation apparatus similarity calculation means calculates the similarity based on each evaluation result.
An output result history storage means for storing the history of the output results;
The integrated evaluation apparatus according to claim 1, wherein the evaluation apparatus similarity calculation unit calculates the evaluation result based on a history of each evaluation result stored in the history storage unit.
A similarity history storage means for storing a history of similarity between the evaluation devices;
2. The integrated evaluation apparatus according to claim 1, wherein the weight setting unit further considers the similarity between the evaluation apparatuses stored in the similarity history storage unit.
2. The integrated evaluation apparatus according to claim 1, wherein the evaluation apparatus similarity calculation means calculates the similarity based on an evaluation result of each evaluation object of the evaluation apparatus as a whole.
The integrated evaluation apparatus according to claim 1, wherein the evaluation apparatus similarity calculation means calculates a similarity based on a range of each evaluation object of the evaluation apparatus.
User action acquisition means for acquiring user action information for an integrated evaluation result created by integrating evaluation results by the integration means;
User behavior history storage means for storing the acquired user behavior information,
The integration means further considers the stored user behavior information history when the user behavior information history of the user who made the evaluation request under the evaluation condition is stored in the user behavior history storage means, The integrated evaluation apparatus according to claim 1, wherein the evaluation results are integrated.
An integrated evaluation system comprising: the integrated evaluation device according to any one of claims 1 to 6, a plurality of evaluation devices, and a user terminal.
An integrated evaluation device according to claim 7, a plurality of evaluation devices, and a user terminal,
The said user terminal is provided with the user action input means which sends the information of the user action performed with respect to the integrated evaluation result acquired from the said integrated evaluation apparatus to the said integrated evaluation apparatus, The integrated evaluation system characterized by the above-mentioned.
Requesting a plurality of evaluation devices for evaluation under evaluation conditions input from a user terminal;
Obtaining an evaluation result corresponding to the request from the plurality of evaluation devices;
Calculating a similarity between the plurality of evaluation devices;
Taking into account the calculated similarity, pressing the weight of the evaluation device having a high similarity, and setting the weight of each evaluation device;
Integrating the evaluation results in consideration of the set weight, and an integrated evaluation method.
A process of requesting evaluations from a plurality of evaluation devices under evaluation conditions input from a user terminal;
Processing for obtaining an evaluation result corresponding to the request from the plurality of evaluation devices;
Processing for calculating the similarity between the plurality of evaluation devices;
In consideration of the calculated similarity, processing to set the weight of each evaluation device by pressing the weight of the evaluation device having a high similarity,
And a process of integrating the evaluation results in consideration of the set weights.
A computer-readable recording medium for recording the processing of the integrated evaluation program according to claim 11.