JP6607691B2 - Evaluation value calculation device and program - Google Patents

Description

  The present invention relates to an evaluation value calculation device and a program.

  In recent years, various users have been engaged in activities such as sharing various kinds of information such as photographs, moving images, and documents, and buying and selling products through communication services using the Internet and the like.

  Among these communication services, search for tag information, etc. in addition to titles, product names, uploaded user information, etc. for photos and products, etc., so that users can more easily find the desired information. Some character string information can be freely registered and stored in association with a target photograph or product.

  As described above, a method for supporting the classification of information by freely tagging the public is called folksonomy and has been widely used in recent years.

  For example, in a photo sharing site, a user who wants to search for a picture of an owl does not search for a title containing “Owl” in the title, but by searching for a tag with the string “Owl”, for example, “ Even if “owl” is not included, a desired photo can be searched.

  In addition, for a product, when the user seeking a home appliance with low power consumption performs a search with a keyword “low power consumption”, for example, when the user searches for a home appliance with low power consumption, the product name is “low power consumption”. Even if the wording is not included, a product including the wording “low power consumption” in the description can be searched.

A. Hotho, et. Al., "Information retrieval in folksonomies: Search and ranking," The Semantic Web: Research and Applications, pp.411-426, 2006

  However, in the conventional communication service described above, the tags and explanations are changed even though the ease of searching for photos and products changes depending on the words included in the tags and explanations, or the degree of interest of the user changes. Currently, it is up to the registrant to provide information on the use of words that are more effective, that is, can increase the number of browsing.

  Non-Patent Document 1 analyzes the user relationship and tagging status regarding resources such as images, and uses a method similar to page rank (a ranking based on a criterion that a higher number of links is more important). Attempts to determine the importance of users and tags have been disclosed.

  However, in the method of Non-Patent Document 1, there is a high probability that a tag such as “bird”, which is a broader concept for an “owl” image or the like, is more important (“bird” is given). This is because the number of resources is generally considered to be greater than the number of “owls”). As described above, in the method of Non-Patent Document 1, a tag or the like that can increase the number of browsing is not necessarily evaluated, and it is not evaluated as a tag or the like that should be given to increase the number of browsing. was there.

  The present invention has been made in view of the above circumstances, and an evaluation value calculation device and a program that can evaluate information elements such as tags and explanations from the viewpoint of effectiveness and can support the provision of effective tags and explanations. One of its purposes is to provide it.

  An evaluation value computing device for solving the problems of the conventional example is connected to a holding unit that holds a plurality of contents set to be distributable via an information communication unit and a point value related to each content. And a point value acquisition means for acquiring a point value for each of the plurality of contents, and a predetermined information element relating to each of the plurality of contents as a processing target, an information element of a processing target obtained from each content, (1) Initializing each element of the vector P in which elements representing evaluation values for each of the listed information elements are arranged to a predetermined value, and using each of the listed information elements as a target element , The value of each column of the adjacency matrix A for the attention element of the undirected graph in which the attention element and the content information including the attention element are linked. A point value of content is determined as a weight, a probability matrix tA is determined so that the sum of elements becomes a predetermined value for each column of the adjacency matrix A, and (2) the probability matrix tA is multiplied by the vector P from the right side. The vector P ′ is updated with the obtained vector P ′, (3) the process of (2) is repeated until the vector P satisfies a predetermined convergence condition, and (4) when the predetermined convergence condition is satisfied. Evaluation value calculation means for outputting the value of each element of the vector P as the evaluation value of the corresponding information element.

  In addition, the evaluation value calculation device according to another aspect of the present invention is connected to a holding unit that holds a plurality of contents set to be distributable via an information communication unit and a point value related to each content, and is accessible. A point value acquisition means for acquiring a point value for each of the plurality of contents, a predetermined information element related to each of the plurality of contents as a processing target, and information elements to be processed obtained from the respective contents are duplicated (1) The element corresponding to the information element related to the content of interest is the first predetermined value, and the element corresponding to the information element not included is the second predetermined value. Formula y = w · x (where w is a weight vector) for each content of interest using vector x and point value y related to the content of interest “·” Represents an inner product), (2) the weight vector is obtained by regression analysis based on the mathematical formula obtained for each content, and (3) the value of the obtained weight vector, Evaluation value calculation means for calculating and outputting an evaluation value of the information element based on the value of the appearance frequency in the plurality of contents for each information element.

  According to the present invention, information elements such as tags and explanatory texts can be evaluated from the viewpoint of effectiveness, and the provision of effective tags and explanatory texts can be supported.

1 is a configuration block diagram illustrating an evaluation value calculation device according to an embodiment of the present invention and a connection example thereof. It is explanatory drawing showing the holding example of the information of the point value in the evaluation value calculating apparatus which concerns on embodiment of this invention. It is a functional block diagram showing the example of the evaluation value calculating apparatus which concerns on embodiment of this invention. It is a flowchart figure showing the operation example of the evaluation value calculating apparatus which concerns on embodiment of this invention. It is explanatory drawing showing the example of the virtual graph used for the calculation by the evaluation value calculating apparatus which concerns on embodiment of this invention. It is a functional block diagram showing another example of the evaluation value calculating apparatus which concerns on embodiment of this invention.

  Embodiments of the present invention will be described with reference to the drawings. As illustrated in FIG. 1, the evaluation value calculation apparatus 1 according to the embodiment of the present invention includes a control unit 11, a storage unit 12, an operation unit 13, a display unit 14, and a communication unit 15. Or the like via a communication device such as a server 2 (a plurality of servers may be connected).

  Here, as illustrated in FIG. 2, the server 2 has a plurality of contents (C) set to be distributable via information communication means such as a network, and a point value (P) regarding each of the plurality of contents. In addition, user specification information (U) for specifying a user related to the content (C) (a user who has registered the content or a user who is a contact or ordering user for the content) is stored in association with each other. Here, the content (C) is, for example, resource information (R), which is the main information such as media data such as photographs and music, and character string information (tag information and description) to be searched in relation to the content information. Sentence information) etc. (G). Here, words, tags, etc. (G) included in the character string information correspond to an example of information elements of the present invention. Here, it is assumed that the tag (G), which is an information element, is included in the content (C). However, if the tag or the like is stored in association with the content (C) (the content is related to the content (C)). It is not necessarily included in the content (C).

  The point value (P) corresponds to an evaluation value for each content (C), and may be, for example, a total score of evaluation points arbitrarily given by a viewer, or may be the number of browsing. .

  In the present embodiment, the server 2 is a web server, and the content (C) is identified by reference information (URL: Uniform Resource Locator).

  The control unit 11 is a program control device such as a CPU, and operates according to a program stored in the storage unit 12. In the present embodiment, the control unit 11 executes processing for calculating the evaluation value of the information element. Details of the processing of the control unit 11 will be described later.

  The storage unit 12 is a memory device or the like and holds a program executed by the control unit 11. This program may be provided by a computer-readable and non-transitory recording medium such as a DVD-ROM and copied to the storage unit 12 or via a communication line such as a network. Provided and copied to the storage unit 12. The storage unit 12 also operates as a work memory for the control unit 11.

  The operation unit 13 is a mouse, a keyboard, or the like. The operation unit 13 accepts a user instruction operation and outputs information representing the content of the accepted instruction operation to the control unit 11. The display unit 14 is a display or the like, and displays information according to an instruction input from the control unit 11.

  The communication unit 15 is connected to information communication means such as a network. The communication unit 15 receives information via a network or the like and outputs the received information to the control unit 11. The communication unit 15 transmits the instructed information (including an information acquisition request) via the network in accordance with an instruction input from the control unit 11.

Here, the operation of the control unit 11 will be described. In the present embodiment, as a method of calculating the evaluation value of the information element,
Since the evaluation value of the information element can be calculated by two methods, (1) a method based on association and (2) a method based on word importance, these will be described separately below.

Association-based method
As illustrated in FIG. 3, the control unit 11 according to the present embodiment that calculates the evaluation value of the information element by the association-based method executes the program stored in the storage unit 12, thereby performing the point value acquisition unit 21. The information element listing unit 22, the iterative calculation unit 23, the convergence determination unit 24, and the evaluation value output unit 25 are functionally realized.

  The point value acquisition unit 21 acquires a point value for each content (C) stored in the server 2. The point value acquisition unit 21 associates information for identifying each content (for example, the URL thereof) and the point value (acquired value) held by the server 2 in association with the content identified by the information. And stored in the storage unit 12 as a point value database.

  For each content (C) stored in the server 2, the information element listing unit 22 acquires a predetermined information element included in the content. Specifically, here, the character string information (description, tag, etc.) to be searched included in the content is acquired. Here, if the acquired character string information is a tag, the information element listing unit 22 performs the following processing for each tag as a processing target. Moreover, if the acquired character string information is a natural sentence such as an explanatory sentence, it is divided into words using a predetermined method (a widely known morphological analysis or the like), and each word obtained by the division is processed. The following processing is performed.

  The information element enumeration unit 22 enumerates information elements obtained from each content, which are processing targets, without duplication, and generates an information element string. Specifically, when a tag such as “bird” or “owl” is included for a certain content α and a tag such as “bird” or “fly” is included for another content γ, the duplication obtained from these is removed. The result of enumerating the information elements (tags) is a set of information elements including three elements “bird”, “owl”, and “fly”. It should be noted that here, duplication is subject to complete matching, and “bird” and “large bird” may be listed as different information elements, partially matched (for example, nouns match), or a thesaurus prepared in advance. Information elements that are judged to represent the same concept may be treated as duplicates (in this case, one of the information elements representing the same concept is used as a representative for information elements of the same concept) Is replaced with the representative information element and the following processing is performed).

  The control unit 11 performs the processing illustrated in FIG. 4 as the operation of the iterative calculation unit 23 and the convergence determination unit 24. First, the iterative calculation unit 23 represents an evaluation value w i (i = 1, 2,..., N) for each of the information elements E i (i = 1, 2,..., N) enumerated by the information element listing unit 22. A vector P in which elements are arranged is generated, and each element is initialized to a predetermined value (for example, “1”). That is, wi = 1 (i = 1, 2,..., N) is set (S1).

となる。この隣接行列Ａにおいては、各行は列挙された各情報要素に対応し、各列はコンテンツに対応する。そしてこの隣接行列Ａは、「情報要素がコンテンツに含まれる」リンクについての隣接行列であり、ある情報要素がコンテンツに含まれる場合に対応する要素を「１」、含まれない場合に対応する要素を「０」としている。繰り返し演算部２３は、この隣接行列Ａの各列の値に、対応するコンテンツについてポイント値取得部２１が得ているポイント値の重みを乗じる（Ｓ４）。 The iterative calculation unit 23 generates an undirected graph in which each of the information elements enumerated by the information element listing unit 22 is used as a target element, and the target element and the content including the target element are linked as nodes (S2). An adjacency matrix A is obtained for each information element of the directed graph (S3). Specifically, when the contents α, γ, and the contents β including the tags “bird”, “owl”, and “fly” are added to the contents α, γ described above, the contents α, β, γ are exemplified as shown in FIG. A direction graph is obtained. In FIG. 5, nodes of information elements are shown as circles and nodes of contents are shown as rectangles. The adjacency matrix A corresponding to the example of FIG.

It becomes. In this adjacency matrix A, each row corresponds to each listed information element, and each column corresponds to content. This adjacency matrix A is an adjacency matrix for the link “information element is included in content”, “1” is an element corresponding to a case where an information element is included in the content, and an element corresponding to a case where it is not included Is set to “0”. The iterative calculation unit 23 multiplies the value of each column of the adjacency matrix A by the weight of the point value obtained by the point value acquisition unit 21 for the corresponding content (S4).

なお、ここでの要素の値は、行方向に要素の数で正規化したもので、例えば第１行目について要素の数が、

となるので、当該行の各値にこの値１／２を乗じて正規化したものである。
Then, the iterative operation unit 23 obtains a probability matrix tA that is determined so that the sum of elements becomes a predetermined value (here, “1”) for each column of the adjacency matrix A after being multiplied by the weight of this point (S5). Specifically, the probability matrix tA is as follows when the point values acquired corresponding to the contents α, β, and γ are 40, 7, and 3, respectively.

The value of the elements of this case is obtained by normalizing the row direction by the number of elements, for example, the number of elements for the first row,

Therefore, each value of the row is normalized by multiplying this value ½.

となる。 This matrix tA can be calculated using the next row probability matrix Bw and column probability matrix Bt. These matrices Bw and Bt are N × M (where N is the number of information elements and M is the number of contents), and the i-th row vector of the matrix Bw is the i-th row vector. It is set so that the point value of the content to which the information element is added is normalized (calculated so that the sum of the point values in the row is “1”), and the content α as in the above example If the point values acquired corresponding to each of, β, and γ are 40, 7, and 3,

It becomes.

となる。 Bt is 1 / s when the i-th element of the j-th column vector is attached with the i-th information element, and 0 when it is not attached. . Here, s is the number of information elements attached to the j-th content, and therefore the sum of the elements of the j-th column vector of this matrix Bt is “1”. Specifically, in the above example, Bt is

It becomes.

The iterative calculation unit 23 calculates the matrix tA as tA = Bw × Bt ^T. Bt ^T means a transposed matrix of the matrix Bt.

  The iterative calculation unit 23 starts the iterative calculation (S6: start of loop), and updates the vector P with a vector P ′ obtained by multiplying the probability matrix tA by the vector P from the right side (S7).

として、第ｎ世代目のベクトルＰnを求めていく。ここでＩは、ベクトルＰと同じ次元で、各要素が「１」であるようなベクトルである。またａは実験的あるいは経験的方法により定められるダンピングファクターである。ダンピングファクターとその設定の方法は、ページランクの演算における同種の方法で広く知られているので、ここでの詳しい説明を省略する。 That means

Then, the nth generation vector Pn is obtained. Here, I is a vector having the same dimension as the vector P and each element being “1”. A is a damping factor determined by an experimental or empirical method. The damping factor and its setting method are widely known as the same kind of method for calculating the page rank, and thus detailed description thereof is omitted here.

  Next, the convergence determination unit 24 determines whether or not the vector Pn has converged based on whether or not a predetermined convergence condition is satisfied (S8). Here, the convergence condition may be a condition such that the norm (which may be the Euclidean norm) of the difference between the vector Pn and the vector Pn−1 is less than a predetermined threshold. When determining that the vector Pn has converged, the convergence determination unit 24 outputs the value of the vector Pn at that time to the evaluation value output unit 25 (S9). If the convergence determination unit 24 determines that the convergence condition is not satisfied in step S8, the convergence determination unit 24 causes the iterative calculation unit 23 to repeat the process of updating the vector P (processing S7) (loop).

  Specifically, in the case of the above example, assuming that the damping factor a = 0.15, the elements of the vector P are 1.28, 1.13, 0.58 in the first update, and converge to 1.07, 1.00, 0.93 after repeated operations.

  The evaluation value output unit 25 obtains the value of the vector P output from the convergence determination unit 24 and performs display based on this value. In this vector P, the value of each element is a value representing the evaluation value of each information element. Therefore, for example, the evaluation value output unit 25 rearranges the values of the elements in descending order (from the largest to the smallest), and displays the information elements corresponding to the element values up to the upper predetermined number as the information elements of the high evaluation value. It is good to do.

  In the above example, the value corresponding to “Bird” is 1.07, the value corresponding to “Owl” is 1.00, and the value corresponding to “Fly” is 0.93. For example, the top two “Birds” and “Owls” As a valid information element.

[Method based on word importance]
Moreover, the control part 11 of this Embodiment may calculate the evaluation value of an information element based on the importance of a word. As illustrated in FIG. 6, the control unit 11 in this example executes a program stored in the storage unit 12, thereby obtaining a point value acquisition unit 21, an information element listing unit 22, a regression calculation unit 31, The appearance frequency calculation unit 32, the evaluation value calculation unit 33, and the evaluation value output unit 25 are functionally realized. In addition, about the thing which has the structure similar to what was already demonstrated, the same code | symbol is attached | subjected and description is simplified.

  The point value acquisition unit 21 acquires a point value for each content (C) stored in the server 2. The point value acquisition unit 21 associates information for identifying each content (for example, the URL thereof) and the point value (acquired value) held by the server 2 in association with the content identified by the information. And stored in the storage unit 12 as a point value database.

  For each content (C) stored in the server 2, the information element listing unit 22 acquires a predetermined information element included in the content. Specifically, here, the character string information (description, tag, etc.) to be searched included in the content is acquired. Here, if the acquired character string information is a tag, the information element listing unit 22 performs the following processing for each tag as a processing target. Moreover, if the acquired character string information is a natural sentence such as an explanatory sentence, it is divided into words using a predetermined method (a widely known morphological analysis or the like), and each word obtained by the division is processed. As described above, information elements obtained from the respective contents and listed as processing targets are listed without duplication to generate an information element string. In the following description, the number of information elements included in the generated information element sequence is N.

  The regression calculation unit 31 sets any one of the contents j (j = 1, 2,..., M) as the attention content jn, and sets an element corresponding to the information element included in the attention content jn to a first predetermined value (for example, “1”). )), An N-dimensional feature vector xjn having elements corresponding to information elements not included as a second predetermined value (for example, “0”), and point values yjn (jn = 1, 2,..., M for the content of interest j). ) To obtain a mathematical expression yjn = w · xjn (where w = (w 1, w 2,..., WN) is a weight vector and “·” represents an inner product) for each content of interest. Then, the regression calculation unit 31 obtains the weight vector w by regression analysis processing based on the mathematical formula yj = w · xj (j = 1, 2,..., M) obtained for each content. Since this regression analysis process can use a widely known process such as SVR (Support Vector Regression), a detailed description thereof is omitted here.

The appearance frequency calculation unit 32 generates an appearance frequency Freq [t, j] (t = 1, 2,... N, j = 1, for each content of each information element included in the information element sequence generated by the information element listing unit 22. 2,... M) is calculated and the sum is obtained. That is, the appearance frequency calculation unit 32 generates an appearance frequency Freq [t] = ΣFreq [t, j] (t = 1, 2,...) In all contents of each information element t (t = 1, 2,... N). N)
(Where Σ means summing over j).

  For each component of the weight vector w = (w 1, w 2,..., W N) obtained by the regression calculation unit 31, the evaluation value calculation unit 33 is information on the appearance frequency of the information element corresponding to the component (appearance frequency calculation unit 32. And the evaluation value Vp for the information element p (p = 1, 2,... N) is Vp = wp × Freq [p] (p = 1, 2,... N). Find and output.

  The evaluation value output unit 25 obtains an evaluation value for the information element p (p = 1, 2,... N) output by the evaluation value calculation unit 33 and performs display based on the evaluation value. For example, the evaluation value output unit 25 rearranges the evaluation values Vp in descending order (in order from large to small), and displays the information elements p corresponding to the evaluation values Vp up to the upper predetermined number as information elements of high evaluation values. It is good to do.

[Operation example]
The evaluation value calculation device 1 of the present embodiment has the above configuration and operates as follows. That is, the evaluation value calculation apparatus 1 according to the present embodiment accepts designation of a data set to be processed from the user. The evaluation value calculation apparatus 1 according to the present embodiment selects data including data set content and point information (information such as evaluation points and the number of views) given to the content as a data set.

  This designation is performed by inputting a list of URLs of contents to be processed or URLs from which the list can be obtained. Specifically, in the flickr (registered trademark), which is a photo sharing site, a list of URLs of content uploaded during a predetermined period can be obtained using the flickr.phots.search API. In such a site, the evaluation value calculation apparatus 1 according to the present embodiment obtains a list of content URLs using an API (Application Program Interface) defined for each site.

  In addition, the evaluation value calculation device 1 acquires at least text or the like (for example, a tag) that is an information element to be processed from each of a plurality of contents specified by URLs included in the obtained list, Get content point information. As an example, the tag is acquired as an information element to be processed here. However, when text including multiple words such as explanatory text is processed, it is divided into words by morphological analysis processing as described above. Then, each word obtained by the division may be acquired as an information element to be processed.

  The evaluation value calculation device 1 uses the content point information to obtain the information element to be processed by either the (1) method based on association or (2) the method based on word importance The evaluation value for each tag is calculated. Then, the calculated evaluation values are rearranged, for example, in descending order, and tags corresponding to the evaluation values up to the upper predetermined number are displayed as high evaluation value information elements.

  Thus, the user can obtain a list of tags from which it is easy to obtain relatively high point information (high evaluation or a large number of browsing). Then, the user can know what kind of tag is included in the content so that a higher evaluation can be obtained or the number of browsing can be increased.

[Guidance of information elements to be associated]
Moreover, the evaluation value calculation apparatus 1 of this Embodiment may perform the process which guides the information element judged that the point information of a content can be improved by including it in object content, such as the content designated by the user. .

  Specifically, the evaluation value calculation apparatus 1 in this example accepts designation of target content from the user together with designation of a data set to be processed. Here, the target content can be specified by specifying the URL. The evaluation value calculation device 1 obtains a target information element group in which predetermined information elements included in the target content are listed from the specified URL without duplication.

  The evaluation value computing device 1 includes at least information elements to be processed from each of a plurality of contents (which may include target contents; hereinafter referred to as data set contents for distinction) that are data sets to be processed. And the like (for example, a tag) and the point information related to each data set content. Since this processing is the same as the example already described, repeated description is omitted.

  The evaluation value calculation device 1 calculates the evaluation value for each information element to be processed by the method (2) based on the importance of the word. Further, the evaluation value calculation apparatus 1 generates a list (comparison information element group) excluding duplication of information elements included in each data set content for each of the plurality of data set contents. Then, the evaluation value calculation device 1 refers to the target information element group enumerated from the target content, compares the target information element group with the comparison information element group related to each data set content, and information related to the correlation is obtained. Then, at least one comparison information element group related to the data set content representing a correlation higher than a predetermined criterion is extracted (collaborative filtering process).

  Specifically, the information relating to the correlation here may be the number of information elements included in common in the target information element group and the comparison information element group. For example, if the information elements included in the target information element group are {x1, x2, x3, x4} and the information elements included in a certain comparison information element group are {x1, x3, x5, x7}, they are common to these. The information elements included are {x1, x3}, and the information related to these correlations is the number “2” of the information elements included in common. In this case, the reference may be a threshold value determined based on the number of information elements included in the target information element group (for example, r% of the number of information elements included in the target information element group).

  The evaluation value computing device 1 enumerates information elements included in the comparison information element group extracted in the collaborative filtering process without duplication, and excludes information elements included in the target information element group from the enumerated information elements, and candidate information elements A group. As a result, information elements extracted from the comparison information element group including information elements similar to the information elements included in the target information element group and not included in the target information element group are obtained. .

  If the candidate information element group is empty, the evaluation value calculation device 1 may display that there is no information element that can be recommended and end the process. If the candidate information element group is not empty, the evaluation value (information calculated previously) for the information element included in the candidate information element group is referred to and the evaluation value exceeds a predetermined threshold value. Present the element. Here, among information elements whose evaluation value exceeds a predetermined threshold value, a predetermined number of information elements having higher evaluation values may be listed and presented.

  Further, the evaluation value calculation apparatus 1 may perform a process of guiding an information element that is determined to be able to improve the point information of the content by using an information element evaluation method based on association without performing a collaborative filtering process.

  The evaluation value calculation device 1 of this example performs the processing illustrated in FIG. 4, but instead of the processing S1, information elements Ei (i = 1, 2,...) Extracted from each data set content and listed. N) generates a vector P in which elements representing evaluation values w i (i = 1, 2,..., N) are arranged, and corresponds to the information elements included in the target information element group among the elements. An element of the vector P is initialized to a first predetermined value (for example, “0”), and an element of the vector P corresponding to an information element not included in the target information element group is set to a second value different from the first predetermined value. The difference is that it is initialized to a predetermined value (eg, “1”). The evaluation value calculation apparatus 1 of this example executes the processes after the process S2 in the same manner as the example illustrated in FIG.

  In this example, the evaluation value calculation device 1 calculates an evaluation value for an information element that is not included in the target information element group. Then, the evaluation value calculation device 1 performs display based on the calculated evaluation value. For example, the evaluation values may be rearranged in descending order (from the largest to the smallest), and information elements corresponding to the upper predetermined number of evaluation values may be displayed as information elements to be included in the target content.

[Negative evaluation]
Moreover, the evaluation value calculation device 1 according to the present embodiment not only presents the evaluation values for each information element in descending order (in descending order of evaluation), but also indicates that the evaluation is lower than a predetermined threshold value. An information element corresponding to the evaluation value to be represented may be presented. According to this, based on the information of the evaluation value for each information element, the information element to be avoided is guided.

[Example of application to explanatory text]
In addition, as described in the above description, the information element may be a word extracted by a process such as morphological analysis from not only a tag but also an explanatory text. In this case, the data set is the content of a web page that sells a predetermined type of product, and the point information is the sales record for each content.

  According to this example, a word used in an explanatory text included in a web page (content) with a relatively high sales record is presented, and the user can use the presented information for refining the explanatory text. Become.

[Example of application to information other than character strings]
Further, in the above description, the information element is a tag or a character string such as a word extracted from the explanatory text. However, the information element of the present embodiment is not limited to this. The information element of this embodiment may be any information element as long as it can be encoded and the relationship with the content is determined. For example, the number of photograph images included in the content, the size (the result of classifying the area occupied by the web page in units of 10% (any of 0% to less than 10%, 10% to less than 20%, etc.)) A value may be used as an information element.

DESCRIPTION OF SYMBOLS 1 Evaluation value calculation apparatus, 2 server, 11 Control part, 12 Storage part, 13 Operation part, 14 Display part, 15 Communication part, 21 Point value acquisition part, 22 Information element enumeration part, 23 Repetition calculation part, 24 Convergence determination part , 25 evaluation value output unit, 31 regression calculation unit, 32 appearance frequency calculation unit, 33 evaluation value calculation unit.

Claims (6)

A plurality of contents set to be distributable via the information communication means and a holding means for holding a point value related to each content are connected to be accessible,
Point value acquisition means for acquiring a point value for each of the plurality of contents;
The predetermined information elements related to each of the plurality of contents are set as processing targets, and the processing target information elements obtained from the respective contents are listed without duplication,
(1) Initializing each element of the vector P in which elements representing evaluation values for each of the listed information elements are arranged to a predetermined value,
Each of the listed information elements as a target element,
Each value of the adjacency matrix A for the target element of the undirected graph that links the target element and the content information including the target element is determined with the point value of the content corresponding to each value as a weight, and the value corresponding to the target element A probability matrix tA is determined by determining that the sum of values of the adjacency matrix A is a predetermined value,
(2) Update the vector P by a vector P ′ obtained by multiplying the probability matrix tA by the vector P from the right side,
(3) The process of (2) is repeated until the vector P satisfies a predetermined convergence condition,
(4) evaluation value calculation means for outputting the value of each element of the vector P when the predetermined convergence condition is satisfied as an evaluation value of the corresponding information element;
An evaluation value calculation device including A plurality of contents set to be distributable via the information communication means and a holding means for holding a point value related to each content are connected to be accessible,
Point value acquisition means for acquiring a point value for each of the plurality of contents;
The predetermined information elements related to each of the plurality of contents are set as processing targets, and the processing target information elements obtained from the respective contents are listed without duplication,
Each content is featured content,
(1) A feature vector x having an element corresponding to an information element related to the content of interest as a first predetermined value and an element corresponding to an information element not included as a second predetermined value, and a point value y related to the content of attention To obtain the formula y = w · x (where w is a weight vector and “·” represents an inner product) for each content of interest,
(2) Obtain the weight vector by regression analysis processing based on the mathematical formula obtained for each content,
(3) an evaluation value calculating means for calculating and outputting an evaluation value of the information element based on the value of the obtained weight vector and the value of the appearance frequency in the plurality of contents for each information element;
An evaluation value calculation device including A plurality of contents set to be distributable via the information communication means and a holding means for holding a point value related to each content are connected to be accessible,
Point value acquisition means for acquiring a point value for each of the plurality of contents;
Means for obtaining a target information element group in which the specified content is listed as target content and the predetermined information elements related to the target content are listed excluding duplication;
The predetermined information elements related to each of the plurality of contents are set as processing targets, and the processing target information elements obtained from the respective contents are listed without duplication,
(1) Among the elements of the vector P in which elements representing evaluation values for each of the listed information elements are arranged, the element of the vector P corresponding to the information element related to the target information element group is set to the first predetermined value. And initializing an element of the vector P corresponding to an information element not included in the target information element group to a second predetermined value different from the first predetermined value,
Each of the listed information elements as a target element,
Each value of the adjacency matrix A for the target element of the undirected graph that links the target element and the content information including the target element is determined with the point value of the content corresponding to each value as a weight, and the value corresponding to the target element A probability matrix tA is determined by determining that the sum of values of the adjacency matrix A is a predetermined value,
(2) Update the vector P by a vector P ′ obtained by multiplying the probability matrix tA by the vector P from the right side,
(3) The process of (2) is repeated until the vector P satisfies a predetermined convergence condition,
(4) means for presenting an information element corresponding to an element satisfying a predetermined condition among the elements of the vector P when the predetermined convergence condition is satisfied;
An evaluation value calculation device including The evaluation value calculation device according to claim 1 or 3,
The means for obtaining the probability matrix tA further normalizes each value of the probability matrix tA by multiplying by 1 / s obtained using the number s of information elements attached to each content information corresponding to each value, An evaluation value calculation device that updates the vector P using the normalized probability matrix tA. A computer that is connected so as to be accessible to a holding unit that holds a plurality of contents set to be distributable via an information communication unit and a point value related to each content,
Point value acquisition means for acquiring a point value for each of the plurality of contents;
The predetermined information elements related to each of the plurality of contents are set as processing targets, and the processing target information elements obtained from the respective contents are listed without duplication,
(1) Initializing each element of the vector P in which elements representing evaluation values for each of the listed information elements are arranged to a predetermined value,
Each of the listed information elements as a target element,
Each value of the adjacency matrix A for the target element of the undirected graph that links the target element and the content information including the target element is determined with the point value of the content corresponding to each value as a weight, and the value corresponding to the target element A probability matrix tA is determined by determining that the sum of values of the adjacency matrix A is a predetermined value,
(2) Update the vector P by a vector P ′ obtained by multiplying the probability matrix tA by the vector P from the right side,
(3) The process of (2) is repeated until the vector P satisfies a predetermined convergence condition,
(4) evaluation value calculation means for outputting the value of each element of the vector P when the predetermined convergence condition is satisfied as an evaluation value of the corresponding information element;
Program to function as. A computer that is connected so as to be accessible to a holding unit that holds a plurality of contents set to be distributable via an information communication unit and a point value related to each content,
Point value acquisition means for acquiring a point value for each of the plurality of contents;
The predetermined information elements related to each of the plurality of contents are set as processing targets, and the processing target information elements obtained from the respective contents are listed without duplication,
Each content is featured content,
(1) A feature vector x having an element corresponding to an information element related to the content of interest as a first predetermined value and an element corresponding to an information element not included as a second predetermined value, and a point value y related to the content of attention To obtain the formula y = w · x (where w is a weight vector and “·” represents an inner product) for each content of interest,
(2) Obtain the weight vector by regression analysis processing based on the mathematical formula obtained for each content,
(3) an evaluation value calculating means for calculating and outputting an evaluation value of the information element based on the value of the obtained weight vector and the value of the appearance frequency in the plurality of contents for each information element;
Program to function as.

Images