JP2010128771A - Clustering result display device, method thereof and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To allow a cluster suiting individual user's interest and preference to be displayed at a higher rank and a cluster index to be displayed using a phrase, etc. known by a user when displaying a clustering result. <P>SOLUTION: Clustering is performed in a predetermined manner to classify a document set obtained through searching a network with a search word inputted by a user into a plurality of clusters, and one or more cluster representative elements are extracted for every cluster. Then, in reference to each previously prepared interest element with its interest element score of the user or a profile corresponding to the user, a total interest element score of the each interest element identical with each cluster representative element of the cluster is obtained as a cluster score of the cluster for every cluster; each cluster is assigned with a rank order according to the obtained score, and the clustering result is displayed according the rank order. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a clustering result display apparatus, a method thereof, and a program for displaying a result of clustering a large number of pieces of document information obtained by searching a communication network based on an arbitrary search word.

  Clustering in net information retrieval refers to a technique of constructing a plurality of subsets (clusters) classified according to some similarity of documents from a document set hit by a search using an arbitrary search word. A typical clustering method is clarified in Non-Patent Document 1, for example. Further, Patent Document 1 discloses an invention in which the display order of each cluster obtained as a result of the clustering process is reconfigured based on the score assigned to each retrieved document.

In recent years, several services have been born that present the result of clustering a document set as a search result of a Web search engine using representative words representing documents belonging to each cluster (for example, Non-Patent Document 2, 3). In these services, by selecting a representative word, it is possible to efficiently acquire a subset desired by the user from the entire search document.
International Publication No. 00/075809 Pamphlet Kishida Kazuaki, “Document Clustering Techniques: Literature Review”, [online], Library and Information science, No. 49, 2003, [September 29, 2008 search], Internet <URL: http: // koara. lib.keio.ac.jp/xoonips/modules/xoonips/download.php?file_id=20480> Group Net Co., Ltd., "What is clustering?", [Online], 2008, Group Net Co., Ltd., [Search September 29, 2008], Internet <URL: http://www.groupnet.co.jp /products/velocity/clustering.html> Cuil, "Drilldowns", [online], 2008, Cuil, Inc., [Search September 29, 2008], Internet <URL: http://www.cuil.com/info/features/>

  In the invention disclosed in Patent Document 1 and the conventional services related to Non-Patent Documents 2 and 3, the results are displayed uniformly for the user without taking into account the interests and preferences of individual users.

  FIG. 20A shows an example of search and clustering results for the search term “jaguar”. In this example, a document set consisting of 100 documents is classified into five clusters. According to the conventional service, when a clustering result as shown in FIG. 20 (a) is displayed, as shown in FIG. 20 (b), the documents are simply rearranged in the order of the number of documents constituting the cluster. A representative word appropriately extracted for each cluster is displayed as an index of the cluster as it is. Therefore, the following problems may occur.

(1) The cluster requested by the user is not displayed at the top. Normally, the display order of the cluster is determined by a measure that is unrelated to the user's individual interests and preferences, such as the number of documents included in the cluster. Is displayed. For this reason, when the user is seeking a minor topic or a specialized topic, a cluster including a large number of topics requested by the user is displayed at a lower level due to the small number of documents, and it takes time to search. In addition, it is considered that an inappropriate (different from the cluster desired by the user) cluster is often selected. In such a case, extra search costs such as rework and re-input are generated. The above problems are particularly remarkable when the number of clusters generated as a result of clustering is large.

(2) The representative word of the cluster is unknown to the user Each cluster is displayed with the representative word representing the document belonging to each cluster as an index. If this representative word is an unknown word for the user, the user cannot estimate the contents of the cluster, and it is difficult to select the cluster that the user wants.

  The object of the present invention is to display a cluster suitable for each user's interests and preferences when displaying the clustering result, and to display the cluster index in the user's known phrases, The clustering result display apparatus and method are realized.

  The clustering result display apparatus according to the present invention includes a search unit, a clustering unit, an interest information storage unit, a calculation unit, and a visualization output unit.

  The search unit searches the communication network for a search term input by the user using an arbitrary search engine, and collects a corresponding document set.

  The clustering unit performs clustering on the collected document set by a predetermined method to classify it into a plurality of clusters, extracts one or more representative elements for each cluster, attaches a cluster representative element score, and outputs the result.

  The interest information accumulation unit accumulates one or more interest elements together with the interest element score for each user or profile.

  The calculation unit refers to each interest element and the interest element score of the profile corresponding to the user or the user stored in the interest information storage unit, and the interest element and each representative element of the cluster match. The sum of the products of the interest element score and the cluster representative element score of each element is used as the cluster score of the cluster, which is obtained for each cluster, and ranking is given to each cluster according to the score.

  The visualization output unit visualizes and outputs the rank of each cluster with the representative element having the largest representative element score as the index of the cluster.

  According to the clustering result display apparatus and method of the present invention, when displaying the clustering result, the cluster that suits each user's interests and preferences is displayed at the top, and the cluster index is displayed in the user's known words and phrases can do.

[First Embodiment]
FIG. 1 is a diagram showing a functional configuration example of the clustering result display device 10 of the present invention, and FIG. 2 is an example of a processing flow thereof. The clustering result display device 10 includes a search unit 11, a clustering unit 12, an interest information storage unit 13, a calculation unit 14, and a visualization output unit 15.

  The search unit 11 searches the communication network 2 using an arbitrary search engine 3 for the search word w input from the user terminal 1 via the communication network 2 by the user u (u is a user number), and the corresponding document The set S is collected (S1).

The clustering unit 12 first classifies the collected document set S into a plurality of clusters C _n (n is a cluster number). Then, for each classified cluster C _n , one or more cluster representative elements R _{n, i} (i is a representative element number) are extracted, and C _n and R _{n, i} are output (S2). As the cluster representative element, for example, a representative word, a representative URL, and the like can be considered, but any element may be used as long as it can serve as an index expressing the contents and characteristics of the cluster. The number of cluster representative elements to be extracted is preferably about several due to the property of representative elements. Further, for example, a cluster representative element score to be described later may be extracted that is equal to or greater than an arbitrarily set threshold value.

  As a clustering method and a method for extracting cluster representative elements, any method may be used. For example, STC (Suffix Tree Clustering) described in Non-Patent Document 1 may be used.

FIG. 3 shows an example in which “jaguar” is searched and clustered as the search word w, and clustering representative elements are extracted. In this example, the document set S consisting of 100 documents, classified respectively the number of documents within five clusters of B _n, for example, the cluster C ₂ is the number of documents B ₂ is 26, two clusters typical element ( Representative word) R _{2, i} is extracted (R _2,1 = feline, R _2,2 = leopard).

In the interest information storage unit 13, for each user or profile, one or more interest elements A _{u, j} (u is a user number or a profile number corresponding to the user u, j is an interest element number), and an interest element score SA. Along with _{u, j} , it is accumulated in advance. Here, the type of the element of interest is the same as the type of the cluster representative element. For example, if the representative element is a representative word, it is an interesting word, and if it is a representative URL, the interesting URL is used.

  The user's interest element is a keyword or the like indicating a hobby / preference of each user. On the other hand, profile interest elements are not related to individual users, but are keywords or the like indicating hobbies, preferences, etc. of users belonging to a specific profile such as age group, gender, and address.

  The interest factor score is a score indicating the degree of interest of each interest element in each user or individual profile. For example, the level of interest is expressed as the magnitude of the score.

Any method may be used as the method for obtaining the interest factor and the interest factor score. For example, the method can be obtained by applying the TF-IDF method (for example, see Reference 1) based on the user's Web browsing history.
[Reference 1] Takenobu Tokunaga, “Information Retrieval and Language Processing”, Language and Calculation 5, University of Tokyo Press, November 1999, p. 27-33

FIG. 4 shows an example of interest information for two users. In this example, the magnitude of interest is expressed as the magnitude of the score, and score = 1 indicates that the interest is maximum, and score = 0 indicates that the interest is minimum. For example, for the user 1, six interest elements (interest words) are extracted, and the fourth interest element A _1,4 “Pro-wrestling” has an interest element score SA _1,4 of 0.6, the sixth. since interest elements a _{1, 6} interest element score SA _{1, 6} of the "car" is 0.1, it comes to that the degree of interest "wrestling" from "car" is large.

The calculation unit 14 first refers to each interest element A _{u, j of} the profile corresponding to the user u or the user u stored in the interest information storage unit 13 and its interest element score SA _{u, j} , and the search unit 11 An element in which each cluster representative element R _{n, i} of each extracted cluster C _n matches each interested element A _{u, j} is extracted. Next, for each cluster C _{n ,} the sum of the interested element scores SA _{u, j} of each matching element is obtained, and this is set as the cluster score SC _{u, n} of the cluster C _n for the user u. Then, the rank P _{u of} each cluster assigned according to the cluster score thus obtained is output (S4). If there is no matching element, the rank may be given in order of the cluster number or the number of documents constituting the cluster, for example.

A method for obtaining the cluster rank _Pu for the user 1 based on the clustering result example shown in FIG. 3 and the interest information example shown in FIG. 4 will be described with reference to FIG. First, an element in which each cluster representative element R _{n, i} of each cluster C _n extracted by the search unit 11 matches each interested element A _{1, j} is “car” for the cluster C _{1 and} for C ₃ . Since “MacOS (registered trademark) X”, “Jobs”, and C ₄ are “pro-wrestling”, these elements are extracted. Next _, the sum of the interest element scores SA _{1, j} of each matching element for each cluster C _n is 0.1 (car) for cluster C _{1 and} 0.7 (= 0.5 for C ₃ ). (MacOS X) +0.2 (Jobs)), because it is 0.6 (wrestling) for _{C 4,} they each cluster score SC _{1, 1, SC} 1, 3, and SC _{l, 4.} Since there are no matching elements for clusters C ₂ and C ₅ , SC ₁ , ₂ and SC _{1, 5} are set to 0. The example of FIG. 4 expresses the magnitude of interest as the magnitude of the score. Therefore, if SC _{u, n} is ranked from the largest numerical value, the degree of interest for each cluster C _{n is shown.} The order P _u is assigned in descending order. In _the example of FIG. 5, since it is _{SC 1,3> SC 1,4> SC 1,1} > SC 1,2 = SC 1,5, if grant rank _{P u} in each cluster _{C n} in this order Good. It should be noted that those having the same ranking, such as C ₂ and C ₅ , may be ranked by appropriately defining rules, for example, depending on the size relationship of the number of documents constituting the cluster.

Finally, the visualization output unit 15 visualizes and outputs the rank P _u for the cluster C _{n using} any cluster representative element R _{n, i} in each cluster C _{n as} an index of the cluster C _n. (S4). In the above example, the cluster order P _u is in the order of C ₃ , C ₄ , C ₁ , C ₂ , C ₅ , so that the representative element number i of each cluster is 1 as shown in FIG. 6, for example. By using elements as indexes and simply following them according to the cluster order and arranging them in the order of “MacOS X → Pro Wrestling → Car → Cat Family → Atari (registered trademark)”, the cluster order _Pu can be visualized. Moreover, you may visualize by making display fonts differ according to the order _| rank Pu of a cluster like tag cloud regardless of how to arrange. 7, while arranging the index of each cluster in descending order of the number of documents contained in the conventional cluster as an example visualized by size rank P _u font clusters obtained by the method of the present invention.

When displaying the clustering results, when ranking is performed in descending order of the number of documents included in the cluster as in the prior art, the cluster ranking P _u is in the order of C ₁ , C ₂ , C ₃ , C ₄ , C _5. On the other hand, according to the clustering result display apparatus and method of the present invention, the user's hobbies and preferences are taken into consideration, and the cluster order P _u is in the order of C ₃ , C ₄ , C ₁ , C ₂ , C _5. . For this reason, even when the user is seeking a minor topic or a specialized topic, clusters that match the interests and preferences of individual users can be easily displayed on the top, making it easier to search. In addition, it can be expected to suppress unnecessary actions such as rework and re-input by selecting a cluster different from the cluster desired by the user.

  Further, as a derivative effect, when profile interest elements are accumulated in the interest information accumulation unit 13, it is possible to process and provide information across profiles to the user. For example, if an interest word score is obtained for a certain interested word by gender, normalization of the male and female scores and the respective ratios are obtained, and as shown in FIG. Differences in degrees can be presented. If interest word scores are obtained for each age group, the scores are normalized for each age group, and the respective ratios are obtained. As shown in FIG. The difference in the degree of interest can be presented. By providing such reference information to the user, it is possible to improve convenience when the user performs a search.

[Second Embodiment]
In the first embodiment, the cluster score of each cluster is obtained based only on the interest factor score corresponding to each cluster, and the rank is assigned to each cluster. On the other hand, in the second embodiment, a cluster score is obtained by further adding a cluster representative element score, and a rank is assigned to each cluster according to the score.

  The rank of the clusters determined by the method of the first embodiment reflects the size of the interest element score corresponding to each cluster as it is, but even if the interest element score is large, in the document set constituting the cluster, When the cluster representative element corresponding to the interest element score is small, it is unlikely that the topic the user wants exists in the cluster. For this reason, even if the cluster is displayed at the top, the absolute amount of topics required by the user is reduced, and there may be a case where effects such as efficient search are not obtained. Therefore, by adding a cluster representative element score indicating the content of cluster representative elements in the document set constituting the cluster to the determination of the cluster score, a cluster having a large absolute amount of topics requested by the user can be easily displayed at the top. .

  FIG. 9 shows a functional configuration example of the clustering result display device 20 of the second embodiment. FIG. 10 shows an example of the processing flow. The clustering result display device 20 includes a search unit 11, a clustering unit 22, an interest information storage unit 13, a calculation unit 24, and a visualization output unit 15. That is, since the clustering unit 12 of the clustering result display device 10 of the first embodiment is replaced with the clustering unit 22 and the calculation unit 14 is replaced with the calculation unit 24, the difference will be described here with emphasis.

The clustering unit 22 extracts one or more cluster representative elements R _{n, i} for each cluster C _n classified by the same process as the clustering unit 12, and also uses a cluster representative element score SR _n, for each cluster representative element _{. i} is assigned, and C _n , R _{n, i} and SR _{n, i} are output (S5).

As described above _{, the} cluster representative element score SR _{n, i} is a value indicating the content of the cluster representative element R _{n, i} in the document set constituting the cluster C _n . Any method may be used. For example, as with the clustering method, it may be obtained using STC, or simply obtained by the ratio of how many documents are included in a plurality of documents that constitute a cluster representative element. May be.

In FIG. 11, each cluster representative element R _{n, i} in the clustering result example shown in FIG. 3 is given a cluster representative element score SR _{n, i} obtained by “number of documents including element ÷ total number of documents in cluster”. An example is shown. For example, for cluster 2, the cluster representative element (representative word) R _2,1 “feline family” has a cluster representative element score SR _2,1 of 1.0. In this case, the phrase “feline family” is It is included in all documents in cluster 2. In addition, since the R _2,2 cluster representative element score SR _2,2 of the "leopard" it is 0.6, the phrase "leopard" is to be included in 60% of all documents in cluster 2.

The computing unit 24 first extracts an element in which each cluster representative element R _{n, i} of the cluster C _n matches each interested element A _{u, j of the} user u, similarly to the processing in the computing unit 14. Next, the cluster representative element score SR _{n, i of} each matching element and the interest element score A _{u, j} are integrated to obtain the representative element score SAR _{u, j, n, i} of each element. Then, the sum is set as the cluster score SC _{u, n} of the cluster C _n for the user u, the rank _Pu is assigned to each cluster according to the cluster score thus obtained, and the rank _Pu is output (S6).

The method of obtaining the cluster rank P ₁ for user 1 based on the interest information example shown in clustering results Examples and Figure 4 shown in FIG. 11 will be described with reference to FIG. First, an element in which each cluster representative element R _{n, i} of each cluster C _n extracted by the search unit 11 matches each interesting element A _{1, j} is a car for the cluster C _{1 and} MacOS X for C _3. and Jobs, since for C ₄ is wrestling, extract these. Next, the cluster representative element score SR _{n, i of} each matching element and the interest element score SA _{u, j} are integrated to obtain a representative element score SAR _{u, j, n, i} . Specifically, 0.1 (= 1.0 × 0.1) for “car”, 0.5 (= 1.0 × 0.5) for “MacOS X”, 0 for “Jobs” 0.08 (= 0.4 × 0.2), and “pro-wrestling” is 0.6 (= 1.0 × 0.6). Then, a total of representative element score for each cluster C _n. Specifically, the cluster C ₁ is 0.1 (car), the C ₃ is 0.58 (= 0.5 (MacOS X) +0.08 (jobs)), and the C ₄ is 0.6 (pro wrestling). Therefore, these are designated as cluster scores SC _1,1 , SC _1,3 and SC _1,4 , respectively. Note that there are no matching elements for clusters C ₂ and C ₅ , and the representative element scores are both 0, so SC ₁ , ₂ and SC _{1, 5} are both 0. As a result, since SC _1,4 > SC _1,3 > SC _1,1 > SC _1,2 = SC _1,5 , the order P ₁ may be assigned to each cluster C _n in this order.

  As described above, by assigning the rank of the cluster based on the cluster score obtained by adding the cluster representative element score, the content level of the element in the document set constituting the cluster is taken into consideration when assigning the rank to the cluster. Therefore, a cluster having a large absolute amount of topics requested by the user can be easily displayed at the top.

[Third Embodiment]
In the second embodiment, in obtaining a cluster score, in addition to an interest element score corresponding to each cluster, a score of each cluster representative element is added, so that a cluster having a large absolute amount of topics requested by the user is displayed at the top. It is easy to be done. In the third embodiment, each cluster score is further weighted according to the number of documents included in each cluster, and the rank of the cluster is determined based on the weighted score.

  The cluster score obtained by the method of the second embodiment is a value adjusted in each cluster so that the absolute amount of the topic requested by the user is easily reflected. However, since the number of documents that make up each cluster is different, to rank the clusters more appropriately reflecting the magnitude of the absolute amount of topics that the user is seeking, the number of documents in each cluster Is also desirable. Therefore, in the third embodiment, the cluster score obtained by the method of each of the above embodiments is further weighted according to the number of documents included in the cluster, and is ranked by the weighted score. By ranking in this way, a cluster having a large absolute amount of topics requested by the user can be more easily displayed on the top.

  FIG. 13 shows a functional configuration example of the clustering result display device 30 of the third embodiment. FIG. 14 shows an example of the processing flow. The clustering result display device 30 includes a search unit 11, a clustering unit 32, an interest information storage unit 13, a calculation unit 34, and a visualization output unit 15. That is, since the difference from the clustering result display device of each of the above embodiments is the clustering unit 32 and the calculation unit 34, the difference will be described here. Here, the second embodiment will be described as a basis.

The clustering unit 32 outputs the number of documents B _n of each cluster together with the cluster C _n , the cluster representative element R _{n, i} , and the cluster representative element score SR _{n, i} obtained by the same processing as the clustering unit 22 ( S7).

The calculation unit 34 adds the number of documents B _n of each cluster to the cluster score SC _{u, n} of each cluster C _n of the user u obtained by the same processing as that of the calculation unit 24, and obtained as a result. In accordance with each score SBC _{u, n} , the rank P _u is assigned to each cluster, and the rank P _u is output (S8).

A method for obtaining the cluster rank P ₁ for the user 1 based on the clustering result example shown in FIG. 11 and the interest information example shown in FIG. 4 will be described with reference to FIG. As a result of the processing similar to that of the calculation unit 24, each cluster score SC _{1, n} of the user 1 is SC _1,1 = 0.1, SC _1,2 = 0, SC _1,3 = 0.58, SC _{1, 4} = 0.6 and SC _1,5 = 0. Each of these, when the number of documents _{B n} of each cluster _{C n} integrated each seek _{SBC 1, n, SBC 1,1 =} 4 (= 0.1 × 40), SBC 1,2 = 0, SBC 1 _{, 3} = 12.18 (= 0.58 × 21), SBC _1,4 = 6 (= 0.6 × 10), and SBC _1,5 = 0. As a result, since SBC _1,3 > SBC _1,4 > SBC _1,1 > SBC _1,2 = SBC _1,5 , rank P ₁ may be given to each cluster C _n in this order.

  As described above, each cluster score is further weighted according to the number of documents included in each cluster, and ranked according to the weighted score, so that a cluster having a large absolute amount of topics requested by the user is ranked higher. It can be displayed more easily.

[Fourth Embodiment]
Each cluster is displayed with any one of the cluster representative elements in each cluster as an index. However, if the cluster representative element is an unknown word or the like for the user, the user cannot estimate the contents of the cluster, and it is difficult to select the desired cluster. Therefore, the fourth embodiment has a configuration in which the cluster representative element having the largest representative element score among the cluster representative elements obtained for each cluster during the processing in the second embodiment is used as the index of the cluster. is there.

The representative element score SAR _{u, j, n, i} is a value obtained by integrating the interest element score A _{u, j} of the user u and the cluster representative element score SR _{n, i of the} cluster C _n . It can be said that it is a value reflecting hobbies and preferences, and a value reflecting the content of the cluster representative elements R _{n, i} in the document set constituting the cluster. Therefore, by using the cluster representative element having the highest representative element score as an index, the index is displayed in words and phrases known to the user, and the contents of the cluster are appropriately reflected in the index, thereby improving the search efficiency. be able to.

  FIG. 16 shows a functional configuration example of the clustering result display device 40 of the fourth embodiment. FIG. 17 shows an example of the processing flow. The clustering result display device 40 is a configuration example based on the clustering result display device 20 of the second embodiment, and includes a search unit 11, a clustering unit 22, an interest information storage unit 13, a calculation unit 44, and a visualization output unit 45. Composed. That is, the calculation unit 44 and the visualization output unit 45 are different from the configuration of the second embodiment. The clustering result display device 40 can also incorporate the configuration of the third embodiment.

The calculation unit 44 outputs the rank P _{u of} each cluster obtained by the same processing as the calculation unit 24, and outputs the representative element score SAR _{u, j, n, i} (S9).

Visualization output unit 45, each cluster representative elements R _n obtained for each cluster C _n, in _i, the typical element score SAR _{u, j, n, i} is the largest cluster typical element R _n, the _i as the index of the cluster, and outputs the rank P _u in each cluster were visualized (S10).

  Based on the clustering result example shown in FIG. 11 and the interest information example shown in FIG. 4, a matching element, a representative element score, a cluster score, and a cluster for each user 2 cluster by the same method as in the second embodiment. The result of obtaining the ranking is shown in FIG. Here, clusters with a cluster score of 0 are ranked with the highest number of documents. Based on this, when the result is displayed simply using the cluster representative element having the representative element number i of 1 for each cluster as an index, as shown in FIG. Display order. On the other hand, when the result is displayed using the cluster representative element having the largest representative element score as an index, it is displayed as “game machine → jaguar XJ → feline department → MacOS X → pro wrestling” as shown in FIG. The As described above, according to the method of the fourth embodiment, instead of “Atari” that is not in the user 2's interest word, “game machine” that is the interest word is displayed, and the interest factor score is relative to the interest word. Instead of a low “car”, “Jaguar XJ” having a relatively high interest factor score is displayed.

  As described above, by using the cluster representative element having the highest representative element score as an index, the index is displayed in words and phrases known to the user, and the contents of the cluster are appropriately expressed in the index, thereby improving the search efficiency. Can be achieved.

  The clustering result display apparatus and method of the present invention are not limited to the above-described embodiments, and can be appropriately changed without departing from the present invention. Each process described above may be executed not only in time series according to the order of description, but also in parallel or individually as required by the processing capability of the apparatus that executes the process.

  Further, when the processing functions of the clustering result display apparatus and method of the present invention are realized by a computer, the processing contents of the functions that the clustering result display apparatus and method should have are described by a program. And the processing function in the said clustering result display apparatus and method is implement | achieved on a computer by running this program with a computer. A part of the processing content may be realized by hardware.

The figure which shows the structural example of the clustering result display apparatus of 1st Embodiment. The figure which shows the example of a processing flow of the clustering result display apparatus of 1st Embodiment. The figure which shows the example of the clustering result in the clustering result display apparatus of 1st Embodiment (MacOS, Apple, and Atari are registered trademarks). The figure which shows an example of user interest information (MacBook is a registered trademark). The figure which shows the calculation image which calculates | requires the cluster order | rank of the user 1 by the clustering result display apparatus of 1st Embodiment. The figure which shows the clustering result display image by the clustering result display apparatus of 1st Embodiment. The figure which shows another clustering result display image by the clustering result display apparatus of 1st Embodiment. The figure which shows the provision image of the information processed based on the user profile accumulate | stored in the interest information storage part. The figure which shows the structural example of the clustering result display apparatus of 2nd Embodiment. The figure which shows the example of a processing flow of the clustering result display apparatus of 2nd Embodiment. The figure which shows the example of the clustering result in the clustering result display apparatus of 2nd Embodiment. The figure which shows the calculation image which calculates | requires the cluster order | rank of the user 1 by the clustering result display apparatus of 2nd Embodiment. The figure which shows the structural example of the clustering result display apparatus of 3rd Embodiment. The figure which shows the example of a processing flow of the clustering result display apparatus of 3rd Embodiment. The figure which shows the calculation image which calculates | requires the cluster order | rank of the user 1 by the clustering result display apparatus of 3rd Embodiment. The figure which shows the structural example of the clustering result display apparatus of 4th Embodiment. The figure which shows the example of a processing flow of the clustering result display apparatus of 4th Embodiment. The figure which shows the calculation image which calculates | requires the cluster order | rank and display index of the user 2 with the clustering result display apparatus of 4th Embodiment. The figure which shows the difference of the clustering result display by the clustering result display apparatus of 2nd Embodiment, and the clustering result display by the clustering result display apparatus of 4th Embodiment. The figure which shows the clustering result display image by a prior art.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 User terminal 2 Communication network 3 Search engine 10, 20, 30, 40 Clustering result display apparatus 11 Search part 12, 22, 32 Clustering part 13 Interest information storage part 14, 24, 34, 44 Calculation part 15, 45 Visualization output part

Claims (17)

A search unit that searches a communication network using a search engine for a search term input by a user and collects a corresponding document set;
A clustering unit that performs clustering by a predetermined method and classifies the document set into a plurality of clusters, and extracts and outputs one or more cluster representative elements for each cluster;
An interest information accumulating unit in which one or more elements of interest are accumulated together with the interest element score for each user or profile;
With reference to each interest element of the user or the profile corresponding to the user stored in the interest information storage unit and the interest element score, each interest element and each cluster representative element of the cluster match each element. A calculation unit that calculates the sum of the interest factor scores as the cluster score of the cluster, calculates it for each cluster, and outputs the rank of each cluster assigned according to the score;
A visualization output unit that visualizes and outputs the ranking for each cluster, using any one of the cluster representative elements in each cluster as an index of the respective cluster;
A clustering result display device comprising: In the clustering result display device according to claim 1,
The clustering unit further outputs a cluster representative element score for each cluster representative element,
The calculation unit further adds the cluster representative element score of each element to the interest element score of each element to obtain a representative element score of each element, and sets the sum as the cluster score of the cluster. Characteristic clustering result display device. In the clustering result display device according to claim 2,
The arithmetic unit further outputs the representative element score,
The clustering result display device, wherein the visualization output unit uses, for each cluster, a cluster representative element having the largest representative element score as an index of the cluster. In the clustering result display device according to any one of claims 1 to 3,
The clustering unit further outputs the number of documents constituting each cluster,
A clustering result display characterized in that the arithmetic unit assigns a rank to each cluster according to a score obtained by further adding the number of documents constituting each cluster to the cluster score obtained for each cluster. apparatus. In the clustering result display device according to claim 1,
The clustering result display device, wherein the visualization output unit visualizes the rank by arranging the clusters according to the rank. In the clustering result display device according to claim 1,
The clustering result display device characterized in that the visualization output unit visualizes the rank by making a font for displaying an index of each cluster different according to the rank. In the clustering result display device according to any one of claims 1 to 6,
A clustering result display device, wherein the cluster representative element is a representative word, and the interesting element is an interesting word. The clustering result display device according to any one of claims 1 to 6,
The clustering result display device, wherein the cluster representative element is a representative URL, and the element of interest is an interest URL. A search step for searching for a search term input by a user on a communication network using an arbitrary search engine and collecting a corresponding document set;
A clustering step for classifying the document set into a plurality of clusters by performing clustering according to a predetermined method, and extracting and outputting one or more cluster representative elements for each cluster;
Each cluster representative element of the cluster and the user or each interest element of the profile corresponding to the user match the interest element score of each element as a cluster score of the cluster, and obtain it for each cluster. A calculation step of outputting the rank of each cluster assigned according to the score;
A visualization output step of visualizing and outputting the ranking for each cluster, using any one of the cluster representative elements in each cluster as an index of the respective cluster,
Clustering result display method to execute. In the clustering result display method according to claim 9,
The clustering step further outputs a cluster representative element score for each of the cluster representative elements,
The calculation step further adds the cluster representative element score of each element to the interest element score of each element to obtain a representative element score of each element, and sets the sum as the cluster score of the cluster. Characteristic clustering result display method. In the clustering result display method according to claim 10,
The calculation step further outputs the representative element score,
In the visualization output step, for each cluster, a cluster representative element having the largest representative element score is used as an index of the cluster. The clustering result display method according to any one of claims 9 to 11,
The clustering step further outputs the number of documents constituting each cluster,
A clustering result display characterized in that the calculation step assigns a rank to each cluster according to a score obtained by adding the number of documents constituting each cluster to the cluster score obtained for each cluster. Method. The clustering result display method according to claim 9 to 12,
The method of displaying a clustering result, wherein the visualization output step visualizes the ranking by arranging the clusters according to the ranking. The clustering result display method according to claim 9 to 12,
In the visualization output step, the ranking is visualized by making a font for displaying an index of each cluster different according to the ranking. The clustering result display method according to any one of claims 9 to 14,
A clustering result display method, wherein the cluster representative element is a representative word, and the interesting element is an interesting word. The clustering result display method according to any one of claims 9 to 14,
A clustering result display method, wherein the cluster representative element is a representative URL, and the element of interest is an interest URL.   The program for functioning a computer as an apparatus in any one of Claims 1-8.