JP2006202118A - Attribute evaluation apparatus, method and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an attribute evaluation apparatus, a method and a program capable of flexibly, efficiently, accurately, and dynamically evaluating attributes related to contents. <P>SOLUTION: Mutually perpendicular vectors associated with candidates for the attributes related to the contents are used as bases, and information related to vectors representing a position within a vector space of general nouns extracted from the contents is created. Based on the created information, the attributes related to the contents are evaluated so as to flexibly, efficiently, accurately, and dynamically evaluate the attributes related to the contents. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an attribute evaluation apparatus, an attribute evaluation method, and an attribute evaluation program that evaluate an attribute related to content based on information related to the content, and in particular, flexibly, efficiently, accurately, and dynamically. The present invention relates to an attribute evaluation apparatus, an attribute evaluation method, and an attribute evaluation program that can evaluate attributes related to content.

  Conventionally, in interactive media such as the Internet, an attribute of content accessed by a user is evaluated. The content attribute is a content category classified by the content content.

  Evaluation of content attributes is particularly important when an Internet advertisement is distributed to content published on the Internet. That is, when the content attribute is “sports”, it is effective to distribute Internet advertisements related to “sports”, and therefore it is necessary to accurately and objectively determine the attributes.

  As a content attribute evaluation method, there is a method in which a person browses the content contents one by one and determines the content attributes from the content contents (see, for example, Non-Patent Document 1). If the content is considered to have a plurality of attributes, the plurality of attributes are assigned to the content.

  For example, when the content is related to “sports news”, it is determined that the content has both an attribute “sport” and an attribute “news”, and both attributes are assigned to the content.

  However, this method has a problem that it is difficult to evaluate content attributes accurately, objectively, and efficiently because people subjectively evaluate content attributes.

  For example, when the content of the content is “sports news”, an example is given in which the content attributes are both “sports” and “news” attributes. It is possible to limit only to the attribute “”, or to further add the attribute “media”.

  As described above, when a person subjectively evaluates an attribute of a content, the evaluation is not constant. Therefore, it is difficult to use the evaluated attribute for distribution of an Internet advertisement or the like.

  Therefore, the content attributes are objectively evaluated in consideration of factors such as keywords extracted from text contained in the content by text mining, the appearance frequency of phrases, font size, website link structure, etc. A method for distributing an Internet advertisement to be distributed to a website on which the content is posted has also been developed (see, for example, Non-Patent Document 2).

Yahoo Japan Corporation, “Category Search”, [online], [Search January 5, 2005], Internet <URL: http://howto.yahoo.co.jp/chapters/8/1.html> Google Inc., “AdSense”, [online], [Search January 18, 2005], Internet <URL: https://www.google.com/adsense/premium-overview?hl=en>

  However, the conventional technique represented by Patent Document 2 described above has a problem that it is difficult to flexibly determine content attributes. In other words, the classification of content attributes changes with changes in language and culture, but the method of determining attributes based on keywords extracted from the content has a problem that it is difficult to cope with such changes. It was.

  For example, advertisers requesting distribution of Internet advertisements are not only directly represented attributes such as “sports” and “news”, but also “warmth”, “sharp”, In recent years, there has been a need to classify content into attributes that are expressed sensuously according to the life scene, such as “warming” and “warm”, and to distribute related advertisements.

  In such a case, it is not easy to classify the content into the above-described attributes desired by the advertiser simply by extracting keywords from the text included in the content, and the attribute evaluation of the content is flexible and dynamic. There was a problem that it was difficult to do.

  Therefore, how to flexibly, efficiently, accurately, and dynamically evaluate content attributes has become a very important issue in order to increase the advertising effectiveness of Internet advertising.

  The present invention has been made to solve the above-described problems caused by the prior art, and is an attribute evaluation apparatus capable of flexibly, efficiently, accurately and dynamically evaluating attributes relating to content. It is an object to provide an attribute evaluation method and an attribute evaluation program.

  In order to solve the above-described problems and achieve the object, the present invention is an attribute evaluation apparatus that evaluates an attribute related to content based on information related to the content, and corresponds to each attribute candidate related to the content Vector information generating means for generating information related to a vector representing a position in a vector space of the information related to the content based on the orthogonal vectors attached to each other, and information generated by the vector information generating means And an attribute evaluation unit that evaluates an attribute related to the content based on the content.

  Further, the present invention, in the above invention, further comprises a database that stores information related to the content and information of candidate attributes related to the content in association with each other, and the vector information generation means searches the information related to the content with a search key. And searching for information on attribute candidates related to content from the database, and generating information related to the vector based on information on candidate attributes related to the content obtained as a result of the search.

  Also, in the present invention according to the above invention, the database stores a plurality of different combinations between the information related to the content and the attribute candidates related to the content, and the vector information generating means is designated from the plurality of combinations. Information on attribute candidates related to the content in the combination is searched, and information related to the vector is generated based on the information on the attribute candidates related to the content obtained as a result of the search.

  Also, in the present invention according to the above-mentioned invention, the database further stores information related to the weight of each candidate for the attribute related to the content, and the vector information generating means includes information about the candidate attribute for the content and the weight The information concerning is read from the database, and the information concerning the vector is generated based on the read information.

  Further, the present invention is characterized in that, in the above-mentioned invention, the attribute evaluation means evaluates a plurality of attributes related to the content and a priority of each attribute with respect to a predetermined content.

  Also, in the present invention according to the above invention, the attribute evaluation unit sets a weight of information related to the vector based on an appearance frequency of the information related to the content, and attributes related to the content based on the set weight. It is characterized by evaluating.

  Also, in the present invention according to the above invention, when the number of attribute candidates related to the content increases, the vector information generation unit adds the vector in the vector space whose dimension increases as the number of candidates increases. Such information is generated.

  Also, the present invention is characterized in that, in the above invention, the information relating to the content is metadata including information relating to the content of the content or information extracted from the content.

  Also, in the present invention according to the above-described invention, the vector information generation unit is associated with each attribute candidate related to the second content associated with the first content to be evaluated by the hyperlink or the trackback. Generating information related to a vector representing a position in the vector space of the information related to the second content based on mutually orthogonal vectors, and the attribute evaluation means is generated by the vector information generation means The attribute relating to the first content is evaluated based on the obtained information.

  Also, the present invention is characterized in that, in the above invention, the information related to the content is a search word used for searching the content or a word set with a hyperlink.

  Further, the present invention is an attribute evaluation method for evaluating an attribute related to the content based on information related to the content, based on mutually orthogonal vectors associated with each candidate for the attribute related to the content, And a vector information generating step for generating information relating to a vector representing a position of the information relating to the content in a vector space, and evaluating an attribute relating to the content based on the information generated by the vector information generating step And an attribute evaluation process.

  Also, in the present invention according to the above invention, the vector information generating step may be configured such that the information related to the content is stored in association with the information related to the content and the attribute candidate information related to the content using the information related to the content as a search key. Information regarding the attribute candidate is searched, and information regarding the vector is generated based on the attribute candidate information regarding the content obtained as a result of the search.

  Also, in the present invention according to the above invention, the vector information generation step may include the attribute related to the content in the specified combination from a database storing a plurality of different combinations between the information related to the content and the attribute candidates related to the content. The information on the vector is searched, and the information on the vector is generated based on the information on the candidate attribute for the content obtained as a result of the search.

  Also, in the present invention according to the above-described invention, the vector information generation step includes the attribute candidate information from the database storing the attribute candidate information related to the content and the information related to the weight of each attribute candidate related to the content. And information related to the weight is read out, and information related to the vector is generated based on the read-out information.

  Further, the present invention is characterized in that, in the above-mentioned invention, the attribute evaluation step evaluates a plurality of attributes related to the content and a priority of each attribute for a predetermined content.

  Also, in the present invention according to the above invention, the attribute evaluation step sets a weight of information related to the vector based on an appearance frequency of the information related to the content, and attributes related to the content based on the set weight. It is characterized by evaluating.

  Further, in the present invention according to the above invention, when the number of attribute candidates related to the content increases, the vector information generation step applies the vector in the vector space whose dimension increases in accordance with the increase in the number of candidates. Such information is generated.

  Also, the present invention is characterized in that, in the above invention, the information relating to the content is metadata including information relating to the content of the content or information extracted from the content.

  Also, in the present invention according to the above invention, the vector information generation step is associated with each attribute candidate related to the second content associated with the first content for which the attribute is evaluated by hyperlink or trackback. Generating information related to a vector representing a position in the vector space of information related to the second content based on orthogonal vectors that are orthogonal to each other, and the attribute evaluation step is generated by the vector information generation step The attribute relating to the first content is evaluated based on the obtained information.

  Also, the present invention is characterized in that, in the above invention, the information related to the content is a search word used for searching the content or a word set with a hyperlink.

  Further, the present invention is an attribute evaluation program for evaluating attributes related to the content based on information related to the content, based on mutually orthogonal vectors associated with the respective attribute candidates related to the content, In addition, a vector information generation procedure for generating information related to a vector representing a position of the information related to the content in a vector space, and an attribute related to the content are evaluated based on the information generated by the vector information generation procedure An attribute evaluation procedure is executed by a computer.

  Also, in the present invention according to the above-mentioned invention, the vector information generation procedure may be configured such that the information related to the content is stored in association with the information related to the content and the attribute candidate information related to the content using the information related to the content as a search key. Information regarding the attribute candidate is searched, and information regarding the vector is generated based on the attribute candidate information regarding the content obtained as a result of the search.

  Further, in the present invention according to the above invention, the vector information generation procedure may be configured such that the attribute relating to the content in the specified combination is stored in a database storing a plurality of different combinations between the information relating to the content and the attribute candidates relating to the content The information on the vector is searched, and the information on the vector is generated based on the information on the candidate attribute for the content obtained as a result of the search.

  Also, in the present invention according to the above-described invention, the vector information generation procedure may include information on the attribute candidates from a database storing information on attribute candidates related to the content and information related to weights of each attribute candidate related to the content. And information related to the weight is read out, and information related to the vector is generated based on the read-out information.

  Also, the present invention is characterized in that, in the above-mentioned invention, the attribute evaluation procedure evaluates a plurality of attributes related to the content and a priority of each attribute with respect to a predetermined content.

  Also, in the present invention according to the above-mentioned invention, the attribute evaluation procedure sets a weight of information related to the vector based on an appearance frequency of the information related to the content, and an attribute related to the content based on the set weight It is characterized by evaluating.

  Further, in the present invention according to the above invention, when the number of attribute candidates related to content is increased, the vector information generation procedure is performed on the vector in the vector space whose dimension is increased in accordance with the increase in the number of candidates. Such information is generated.

  Also, the present invention is characterized in that, in the above invention, the information relating to the content is metadata including information relating to the content of the content or information extracted from the content.

  Also, in the present invention according to the above invention, the vector information generation procedure is associated with each attribute candidate related to the second content associated with the first content for which the attribute is evaluated by a hyperlink or a trackback. Generating information related to a vector representing a position in the vector space of the information related to the second content based on the vectors orthogonal to each other, and the attribute evaluation procedure is generated by the vector information generation procedure The attribute relating to the first content is evaluated based on the obtained information.

  Also, the present invention is characterized in that, in the above invention, the information related to the content is a search word used for searching the content or a word set with a hyperlink.

  According to the present invention, information related to a vector representing a position in a vector space of information related to content is generated based on vectors orthogonal to each other associated with each candidate for the attribute related to the content. Since the attribute related to the content is evaluated based on the information, the attribute related to the content can be evaluated flexibly, efficiently, accurately and dynamically.

  Further, according to the present invention, the information on the content is searched for from the database in which the information on the content is stored in association with the information on the content and the information on the attribute candidate on the content using the information on the content as a search key. Since the information related to the vector is generated based on the attribute candidate information related to the content obtained as a result of the search, it is efficient by reading the attribute candidate information related to the content stored in the database. This produces an effect that vector information can be generated.

  Further, according to the present invention, information on attribute candidates related to content in a specified combination is searched from a database storing a plurality of different combinations between information related to content and content attribute candidates. Since the information related to the vector is generated based on the candidate attribute information related to the content obtained as a result of the above, it is possible to flexibly evaluate the attribute related to the content.

  Further, according to the present invention, the information on the candidate attributes and the information on the weights are read from the database storing the information on the attribute candidates related to the contents and the information on the weights of the respective attribute candidates related to the contents. Since the information related to the vector is generated based on the content, the attribute related to the content can be evaluated more accurately by evaluating the information related to the content in consideration of the weight of each candidate for the attribute related to the content. There is an effect that can be.

  Further, according to the present invention, since a plurality of attributes related to the content and the priority of each attribute are evaluated for a predetermined content, the attribute related to the content can be evaluated with arbitrary accuracy. There is an effect.

  Also, according to the present invention, the weight of information related to the vector is set based on the appearance frequency of the information related to the content, and the attribute related to the content is evaluated based on the set weight. By considering the appearance frequency of information, there is an effect that the attribute related to the content can be more accurately evaluated.

  Further, according to the present invention, when the number of attribute candidates related to the content increases, the information related to the vector in the vector space whose dimension increases in accordance with the increase in the number of candidates, There is an effect that it is possible to dynamically cope with an increase in the number of attribute candidates related to the content.

  Further, according to the present invention, since the information related to the content is metadata including information related to the content of the content or information extracted from the content, the information extracted from the metadata or content is Based on this, it is possible to evaluate the attribute relating to the content flexibly, efficiently, accurately and dynamically.

  In addition, according to the present invention, the vectors orthogonal to each other associated with the respective attribute candidates related to the second content associated with the first content to be evaluated by the hyperlink or the trackback are used as a basis. Since the information related to the vector representing the position of the information related to the second content in the vector space is generated and the attribute related to the first content is evaluated based on the generated information, the hyperlink Alternatively, it is possible to evaluate the attribute of the first content flexibly, efficiently, accurately, and dynamically from the second content associated by the track back.

  Further, according to the present invention, since the information related to the content is a search word used for content search or a word set with a hyperlink, the search word used for content search, Alternatively, it is possible to evaluate the attribute relating to the content flexibly, efficiently, accurately and dynamically based on the word for which the hyperlink is set.

  Exemplary embodiments of an attribute evaluation apparatus, an attribute evaluation method, and an attribute evaluation program according to the present invention will be described below in detail with reference to the accompanying drawings.

  First, the concept of attribute evaluation processing according to the present invention will be described. FIG. 1 is a diagram for explaining the concept of attribute evaluation processing according to the present invention. FIG. 1 is an example in the case of evaluating an attribute relating to a content composed of texts such as “I am a graduate student who likes composition with a synthesizer, and major is natural language processing.”

  Here, the attribute relating to the content includes a category to which the content determined from the content content belongs, an attribute of a user accessing the content, and the like. For example, when the content has an attribute of “sports”, the user who accesses the content is evaluated as a user who is interested in “sports”, and “sports” is set as an attribute related to the user's interests. .

  In this attribute evaluation process, first, morphological analysis is first performed on the text. FIG. 2 is a diagram illustrating an example of the execution result of the morphological analysis. As shown in FIG. 2, in the morpheme analysis, the text is decomposed into each morpheme, and a process of analyzing the part of speech of each morpheme is performed.

  Subsequently, a morpheme whose part of speech is a general noun (corresponding to “noun-general” in FIG. 2) is extracted. In the example of FIG. 2, two common nouns “synthesizer” and “natural language processing” are extracted.

  In this example, only common nouns are extracted, but addi- tional connection nouns such as “Graduation”, “Composition”, “Major” (corresponding to “Noun-Sabari Connection” in FIG. 2) are further added. It is good.

  On the other hand, a database storing basic attribute elements corresponding to general nouns and information on ratios between basic attribute elements is prepared in advance. Hereinafter, this database is referred to as a digital thesaurus. FIG. 3 is a diagram illustrating an example of a digital thesaurus that stores information on basic attribute elements and ratios between basic attribute elements.

  As shown in FIG. 3, the digital thesaurus stores information on general nouns, basic attribute elements, and element ratios. A general noun is a word whose part of speech is classified as a general noun. A basic attribute element is a category to which a general noun belongs.

  For example, in the example of FIG. 3, “synthesizer” belongs to two basic attribute elements “music” and “computer”, and “natural language processing” is two basic attribute elements “computer” and “linguistics”. Belongs to.

  The element ratio is a ratio assigned between basic attribute elements. For example, in the example of FIG. 3, the element ratio of two basic attribute elements of “synthesizer” “music” and “computer” is set to 1: 1, and two basic attribute elements of “natural language processing” are set. The element ratio of a certain “computer” and “music” is set to 3: 2. This element ratio is used as a weight for the basic attribute element.

  In this attribute evaluation process, when a general noun is extracted from the content, basic attribute elements and element ratio information corresponding to the general noun are searched from the digital thesaurus. Based on these pieces of information, as shown in FIG. 1, the positions of the general nouns in the vector space (Hilbert space) are expressed using base vectors that correspond to the basic attribute elements and are orthogonal to each other.

Specifically, general nouns are expressed as follows:
n _i = Σa _j | e _j >. . . (Formula 1)
Σa _j ² = 1. . . (Formula 2)

Here, n _i is a unit vector of size 1 representing the general noun i, | e _j > is an orthonormal basis vector of size 1 corresponding to the basic attribute element j, and a _j is , A weight of | e _j > corresponding to the basic attribute element j and calculated from the element ratio, Σa _j | e _j > is the sum of a _j | e _j > for all j, and Σa _j ² Is the sum of a _j ² for all j.

For example, since the basic attribute elements corresponding to the general noun “synthesizer” are “music” and “computer”, and the element ratio of “music” and “computer” is 1: 1, the unit vector representing “synthesizer” n ₁ is
n ₁ = 1 / √2 | e ₁ > + 1 / √2 | e ₂ >. . . (Formula 3)
It becomes.

Here, | e ₁ > is a basis vector corresponding to “music”, and | e ₂ > is a basis vector corresponding to “computer”. Since the element ratio of “music” and “computer” is 1: 1, the weights of | e ₁ > and | e ₂ > are also set to 1: 1. Specifically, from Equation 2 and the element ratio, the weights of | e ₁ > and | e ₂ > are 1 / √2 and 1 / √2, respectively.

The basic attribute elements corresponding to the general noun “natural language processing” are “computer” and “linguistics”, and the ratio of “computer” and “linguistics” is 3: 2. The unit vector n ₂ representing “language processing” is
n ₂ = 3 / √13 | e ₂ > + 2 / √13 | e ₃ >. . . (Formula 4)
It becomes.

Here, | e ₃ > is a basis vector corresponding to “linguistics”. Further, from Equation 2 and element ratio 3: 2, the weights of | e ₂ > and | e ₃ > are 3 / √13 and 2 / √13, respectively.

The attribute related to the content is the sum of vectors corresponding to the respective general nouns, that is,
p = Σn _i . . . (Formula 5)
It is evaluated by. Here, p is an attribute vector representing an attribute related to the content, and Σn _i is the sum of n _i for all i.

For example, an attribute vector p representing attributes related to content from which general nouns “synthesizer” and “natural language processing” are extracted is:
p = n ₁ + n ₂
= 1 / √2 | e ₁ > + (1 / √2 + 3 / √13) | e ₂ > + 2 / √13 | e ₃ >
. . . (Formula 6)
It becomes.

As a result, the percentage of content attributes “music”, “computer”, and “linguistics”
1 / √2: (1 / √2 + 3 / √13): 2 / √13
≒ 25%: 55%: 20%. . . (Formula 7)
It is evaluated.

  Here, the ratio can be considered as a priority in selecting an attribute related to content. That is, if it is necessary to determine one attribute related to the content, “computer” is selected, and if two attributes related to the content are determined, “computer” and “music” may be selected.

Here, as shown in Expression 7, the ratios of the coefficients corresponding to the respective base vectors | e _j > are simply compared. However, the attribute vectors of Expression 5 are normalized, and each of the normalized attribute vectors is The ratio of the squares of the coefficients corresponding to the basis vectors | e _j > may be compared.

For example, when the attribute vector p represented by Equation 6 is normalized,
_{p = 0.3967 | e 1> +} 0.8636 | e 2> + 0.3112 | e 3>
It becomes.

In this case, the percentage of attributes related to the content is “music”, “computer”, “linguistics”
0.3967 ² : 0.8636 ² : 0.3112 ²
≒ 16%: 74%: 10%
It is evaluated.

  Further, if the size of the attribute vector for evaluating the attribute related to the content is normalized, there is an advantage that the coefficient corresponding to the base vector can be compared with the attribute vector of other content.

  Next, the reason why vectors representing general nouns are represented using mutually orthogonal basis vectors will be described. FIG. 4 is a diagram for explaining an attribute evaluation method for content when base vectors that are not orthogonal to each other are used. For example, consider a case where an attribute relating to a content consisting of a text “thinking about quantization encryption while listening to music” is evaluated.

  Then, by the morphological analysis, general nouns “music” and “quantized cipher” are extracted, and “music” is selected as the basic attribute element corresponding to “music”, and “ It is assumed that “physics” and “computer” are searched from the digital thesaurus, and further, element ratios corresponding to the respective basic attribute elements are searched.

FIG. 4 shows vectors n ₃ and n ₄ calculated on the basis of basic attribute elements and element ratios, corresponding to the general nouns “music” and “quantized cipher”. However, the bases | e ₄ > to | e ₇ > of the vectors are not orthogonal to each other.

In such a case, if the sum of the vectors n ₃ and n ₄ is simply calculated based on Equation 5, the resulting attribute vector q does not have the element “music”, and the content is irrelevant to “music”. May be wrong. In order to prevent this, here, vectors that are orthogonal to each other are used as basis vectors of vectors representing general nouns.

  As described above, in the attribute evaluation processing according to the present invention, information related to a vector representing a position in a vector space of a general noun extracted from content based on mutually orthogonal vectors associated with basic attribute elements. And attributes related to the content are evaluated based on the generated information.

  Here, any basic attribute element can be associated with the underlying vector, so Internet advertising advertisers can “warm”, “sharp”, “warm”, “warm” Even when it is desired to newly add a sensually expressed attribute according to the life scene, etc., it can be flexibly handled by updating the digital thesaurus.

  Moreover, since the attribute which concerns on a content is evaluated using the vector which mutually orthogonally crosses, an attribute can be evaluated correctly. Furthermore, since the attribute related to the content is evaluated by vector calculation, the attribute can be efficiently evaluated.

  As described above, when there are N basic attribute elements, the general noun is represented by an N-dimensional vector based on an orthonormal vector corresponding to the basic attribute element. Therefore, even when a basic attribute element for a general noun is added to the digital thesaurus, it is possible to dynamically respond to the addition of the basic attribute element by simply increasing the dimension of the vector.

  Next, a functional configuration of the attribute evaluation system according to the present embodiment will be described. FIG. 5 is a diagram illustrating a functional configuration of the attribute evaluation system 20 according to the present embodiment. Below, the case where the attribute which concerns on the content of the website published via the internet 30 is evaluated is demonstrated.

  As shown in FIG. 5, the attribute evaluation system 20 is connected to external web servers 10 a to 10 c via the Internet 30. The external web servers 10 a to 10 c are servers that provide website browsing services outside the attribute evaluation system 20. The external web servers 10a to 10c store HTML (Hyper Text Markup Language) data for constructing a website and an access log that is a user's access history to the website.

  The attribute evaluation system 20 is a system that provides a content browsing service on a website and evaluates the content of the browsing service and the attributes of a user who accesses the content.

  The attribute evaluation system 20 has a configuration in which web servers 40a to 40c, an information collection server 50, an attribute evaluation server 60, and an advertisement server 70 are connected via a LAN (Local Area Network) 80. The web servers 40a to 40c, the information collection server 50, and the advertisement server 70 are connected to the external web servers 10a to 10c via the Internet 30.

  The web servers 40a to 40c are servers that provide a browsing service for content in a website. The web servers 40a to 40c store HTML data for constructing a website and an access log that is a user's access history to the website.

  The information collection server 50 accesses another server and collects link information set between website contents, metadata information of website contents, contents information of websites, and the like. It is. The information collection server 50 includes a data transmission / reception unit 500, an information collection unit 501, a storage unit 502, and a control unit 503.

  The data transmission / reception unit 500 is a network interface that exchanges various data with other servers via the Internet 30 or the LAN 80. The information collection unit 501 accesses the external web servers 10a to 10c and the web servers 40a to 40c connected to the Internet 30, and information on links set between the contents of the website and the metadata of the contents in the website. Information, information on contents in the website, and the like are collected and stored in the storage unit 502.

  The storage unit 502 is a storage device such as a hard disk device. The storage unit 502 stores link information 502a, metadata information 502b, and content information 502c.

  The link information 502a stores information on links set between contents. This link is set by a hyperlink or trackback. The metadata information 502b stores metadata information describing information related to contents in the website. The content information 502c stores content information such as text and image data in the website.

  The control unit 503 is a control unit that controls the information collection server 50 as a whole, and controls data exchange between the functional units.

  The attribute evaluation server 60 acquires the information collected by the information collection server 50 and performs processing for evaluating the attribute of the content to be evaluated and the attribute of the user accessing the content.

  The attribute evaluation server 60 includes a data transmission / reception unit 600, a storage unit 601, a vector information generation unit 602, an attribute evaluation unit 603, and a control unit 604. The data transmission / reception unit 600 is a network interface that exchanges various data with other servers via the LAN 80.

  The storage unit 601 is a storage device such as a hard disk device. The storage unit 601 stores digital thesaurus data 601a and attribute evaluation information 601b.

  The digital thesaurus data 601a corresponds to the digital thesaurus described with reference to FIG. The digital thesaurus data 601a stores information on general nouns, basic attribute elements, and element ratios. The attribute evaluation information 601b stores information regarding the attribute of the evaluated content and the attribute of the user who accesses the content.

  Here, only one combination between the general noun, the basic attribute element, and the element ratio is stored in the digital thesaurus data 601a. However, the content and the content are accessed by the Internet advertisement advertiser as desired. In order to classify users, a plurality of different combinations among general nouns, basic attribute elements, and element ratios may be stored in the digital thesaurus data 601a, and the combinations to be used may be switched according to the advertiser.

  Here, the combination to be used may be set in advance for each advertiser, or vector information is generated from a combination designated by the advertiser using a terminal device (not shown) connected to the Internet 30. It is good also as setting by the part 602 receiving.

  This allows one advertiser to use directly expressed elements such as “sports” or “news” as basic attribute elements, while other advertisers use “warmth” or “sharp” Even if there is a demand for using elements expressed sensuously, such as “warm and warm” or “warm”, it is possible to respond flexibly.

  Returning to the description of FIG. 5, the vector information generation unit 602 performs morphological analysis on the content and extracts general nouns from the content. Then, the vector information generation unit 602 performs processing for expressing the extracted general noun using base vectors orthogonal to each other and calculating a position in the vector space. Specifically, the vector information generation unit 602 refers to the digital thesaurus data 601a and expresses general nouns using Equations 1 and 2 as described in FIG.

  The attribute evaluation unit 603 performs a process of evaluating the attribute of the content and the attribute of the user who accesses the content based on the position information in the vector space of the general nouns expressed using the mutually orthogonal basis vectors. .

  Specifically, as described with reference to FIG. 1, the attribute evaluation unit 603 generates an attribute vector from the sum of vectors representing general nouns using Equation 5, and calculates the coefficient of each base vector in the attribute vector as Equation 7 The attribute is evaluated by checking as described above.

  The advertisement server 70 is a server that distributes an advertisement suitable for each content and user based on the attribute of the content evaluated by the attribute evaluation server 60 and the attribute of the user who accesses the content. The advertisement server 70 includes a data transmission / reception unit 700, a storage unit 701, an advertisement distribution processing unit 702, and a control unit 703.

  The data transmission / reception unit 700 is a network interface that exchanges various data with other devices via the Internet 30 or the LAN 80. The storage unit 701 is a storage device such as a hard disk device. The storage unit 701 stores advertisement data 701a and distribution condition data 701b.

  The advertisement data 701a stores data of Internet advertisements that are distributed to content provided by the external web servers 10a to 10c or the web servers 40a to 40c. The distribution condition data 701b is data storing distribution conditions for Internet advertisements such as a URI (Uniform Resource Identifier) of contents for distributing Internet advertisements and a distribution period.

  The advertisement distribution processing unit 702 performs processing for distributing the Internet advertisement stored in the advertisement data 701a to the content based on the distribution condition data 701b stored in the storage unit 701. The control unit 703 is a control unit that controls the advertisement server 70 as a whole, and controls data exchange between the functional units.

  Next, a processing procedure of attribute evaluation processing according to the present embodiment will be described. FIG. 6 is a flowchart illustrating a processing procedure of attribute evaluation processing according to the present embodiment.

  As shown in FIG. 6, first, the vector information generation unit 602 of the attribute evaluation server 60 acquires information on the content that is the target of attribute evaluation from the content information 502 c stored in the storage unit 502 of the information collection server 50. (Step S101).

  Then, as described with reference to FIG. 2, the vector information generation unit 602 performs morphological analysis on the content information (step S102), and extracts general nouns (step S103).

  Subsequently, the vector information generation unit 602 refers to the digital thesaurus data 601a stored in the storage unit 601 to search for basic attribute elements and element ratios corresponding to the extracted general nouns (step S104). And the general noun is converted into a vector by using Equation 2 (step S105).

  Then, the attribute evaluation unit 603 calculates the sum of the vectors corresponding to the respective general nouns using Equation 5 (step S106), and uses the coefficient of each base vector in the vector sum as described in Equation 7. The attribute related to the content is evaluated (step S107), the attribute related to the evaluated content is stored in the storage unit 601 as the attribute evaluation information 601b (step S108), and this attribute evaluation process ends.

  In the above embodiment, general nouns are extracted from content, general nouns are converted into vectors based on basic attribute elements, and attributes related to content are evaluated. However, instead of extracting general nouns from content, In addition, when searching for the content using a search engine, a search term input by the user or a hyperlink to the content is set, and the word clicked by the user with a mouse or the like is converted into a vector, and the content The attribute may be evaluated.

  As described above, in this embodiment, the vector information generation unit 602 of the attribute evaluation server 60 uses the vectors orthogonal to each other associated with each basic attribute element as a basis, and in the vector space of general nouns. Since the information related to the vector representing the position is generated and the attribute evaluation unit 603 evaluates the attribute related to the content based on the information generated by the vector information generation unit 602, it is flexible and efficient. It is possible to accurately and dynamically evaluate the attributes relating to the content.

  Further, in this embodiment, the vector information generation unit 602 uses the general noun as a search key to retrieve basic attribute element information from the digital thesaurus data 601a in which the general noun and the attribute candidate information related to the content are stored in association with each other. Since the search is performed and the attribute evaluation unit 603 generates information related to the vector representing the position of the general noun in the vector space based on the information of the basic attribute element obtained as a result of the search, the digital thesaurus By reading the basic attribute element information stored in the data 601a, vector information can be generated efficiently.

  In this embodiment, the vector information generation unit 602 searches the digital thesaurus data 601a that stores a plurality of different combinations between general nouns and basic attribute elements, and searches for information on the basic attribute elements in the specified combination. Since the information related to the vector is generated based on the basic attribute element information obtained as a result of the search, the attribute related to the content can be flexibly evaluated.

  In this embodiment, the vector information generation unit 602 reads basic attribute element information and element ratio information from digital thesaurus data 601a storing basic attribute element information and basic attribute element ratio information. Since the attribute evaluation unit 603 generates information related to the vector representing the position of the general noun in the vector space based on the read information, the information related to the content in consideration of the element ratio of the basic attribute elements By performing the evaluation, it is possible to evaluate the attribute related to the content more accurately.

  In this embodiment, since the attribute evaluation unit 603 evaluates a plurality of attributes related to the content and the priority of each attribute with respect to the predetermined content, the attribute related to the content is determined with arbitrary accuracy. Can be evaluated.

  In this embodiment, when the number of attribute candidates related to the content increases, the vector information generation unit 602 represents the position of the general noun in the vector space whose dimension increases in accordance with the increase in the number of candidates. Since the information related to the vector is generated, it is possible to dynamically cope with an increase in the number of attribute candidates related to the content.

  In this embodiment, the general noun is metadata including information related to the content or extracted from the content. Therefore, the general noun is based on the general noun extracted from the metadata or content. Thus, it is possible to evaluate the attribute relating to the content flexibly, efficiently, accurately and dynamically.

  Further, in this embodiment, the vector information generation unit 602 sets a search word or a hyperlink that is based on vectors orthogonal to each other that are associated with the basic attribute elements and that is used for content search. Since the information related to the vector representing the position of the word in the vector space is generated, it is flexible and efficient based on the search word used for content search or the word set with the hyperlink. Therefore, it is possible to accurately and dynamically evaluate the attribute relating to the content.

(Modification 1 of an Example)
By the way, in the above embodiment, the attribute related to the content is evaluated on the basis of the basic attribute element corresponding to the general noun and the element ratio. Such an attribute may be evaluated. Therefore, in Modification 1 of the embodiment, a case where the appearance frequency of general nouns is further considered will be described.

  Here, the first text used in the above example, “I like to synthesize music with a synthesizer, my major is natural language processing,” and “Natural language processing is The example in the case of evaluating the attribute which concerns on the content which consists of the 2nd text "It is pleasant."

  FIG. 7 is a diagram illustrating an example of the execution result of the morphological analysis in the first modification. FIG. 7 shows the result of applying morphological analysis to the second text. The result of performing the morphological analysis on the first text is the same as that shown in FIG.

  As shown in FIG. 7, one general noun “natural language processing” is extracted from the second text. Thereby, the general nouns extracted from the first text and the second text are “synthesizer” and “natural language processing science”, and the appearance frequencies are once and twice, respectively.

  In the attribute evaluation process according to the first modification, when evaluating an attribute relating to content, the appearance frequency of a general noun is set as a vector weight corresponding to the general noun, and the sum of the vectors is calculated.

That is, in the attribute evaluation process in the first modification, instead of Equation 5,
p = Σw _i n _i . . . (Formula 8)
Is used to calculate the sum of the vectors. Here, w _i is the appearance frequency of the general noun corresponding to the vector n _i , and Σw _i n _i is the sum of w _i n _i for all i.

For example, the general nouns “synthesizer” and “natural language processing” are extracted, the frequency of appearance of “synthesizer” is once, and the attribute vector p of the content where “natural language processing” appears twice Is
p = n ₁ +2 n ₂
= 1 / √2 | e ₁ > + 1 / √2 | e ₂ > +2 (3 / √13 | e ₂ > + 2 / √13 | e ₃ >)
= 1 / √2 | e ₁ > + (1 / √2 + 6 / √13) | e ₂ > + 4 / √13 | e ₃ >
. . . (Formula 9)
It becomes.

As a result, the percentage of content attributes “music”, “computer”, and “linguistics”
1 / √2: (1 / √2 + 6 / √13): 4 / √13
≒ 17%: 57%: 26%. . . (Formula 10)
It becomes.

  Therefore, when it is necessary to determine one attribute related to the content, “computer” is selected, and when two attributes related to the content are determined, “computer” and “linguistics” may be selected. .

  Note that the functional configuration of the attribute evaluation system according to Modification 1 is substantially the same as the functional configuration shown in FIG. However, in the first modification, the vector information generation unit 602 performs morphological analysis on the content, extracts general nouns from the content, and stores the appearance frequency of the extracted general nouns in the storage unit 601.

  Then, as described with reference to FIG. 1, the vector information generation unit 602 represents the extracted general nouns using base vectors that are orthogonal to each other, and performs processing for calculating the positions of the general nouns in the vector space.

  Further, the attribute evaluation unit 603 uses the expression 8 to calculate the content attributes and the general nouns from the appearance frequency of the general nouns and the position information in the vector space of the general nouns expressed using mutually orthogonal basis vectors. Evaluate the attributes of users accessing the content.

  Further, the processing procedure of the attribute evaluation processing according to the first modification is almost the same as the processing procedure shown in FIG. However, in the first modification, in step S103, the vector information generation unit 602 extracts general nouns from the content, and stores the appearance frequency of the extracted general nouns in the storage unit 601.

  In step S106, the attribute evaluation unit 603 calculates the sum of the vectors corresponding to each general noun using Expression 8 instead of Expression 5. In step S107, the attribute evaluation unit 603 evaluates the attribute related to the content from the coefficient of each base vector in the vector sum as described in Expression 10.

  As described above, in the first modification of the present embodiment, the attribute evaluation unit 603 sets and sets the weight of information related to the vector generated by the vector information generation unit 602 based on the appearance frequency of the general noun. Since the attribute relating to the content is evaluated based on the weights thus determined, the attribute relating to the content can be more accurately evaluated by considering the appearance frequency of the general noun.

(Modification 2 of an Example)
By the way, in the first embodiment and the first modification of the embodiment, the content is text. However, even when the content is an image, the metadata set for the image is analyzed and the content is analyzed. Such an attribute may be evaluated. Therefore, in a second modification of the embodiment, a case will be described in which attributes relating to content including an image together with text are evaluated.

  Here, the text used in Modification 1 above, “I graduated, I like to compose with a synthesizer, my major is natural language processing, and natural language processing is fun” and the synthesizer. An example in the case of evaluating an attribute related to a content including a score image and a score is shown.

  Here, the metadata of the synthesizer image includes a general noun “synthesizer”, and the metadata of the score image includes a general noun “music”.

  In this case, when general nouns are extracted from the text and metadata by morphological analysis, the general nouns “synthesizer”, “natural language processing” and “music” are extracted, and the appearance frequency of these general nouns is twice. 2 times and 1 time.

In this case, the attribute vector p representing the attribute related to the content is expressed by
_{p = 2 n 1 + 2 n} 2 + n 3
= 2 (1 / √2 | e ₁ > + 1 / √2 | e ₂ >)
+2 (3 / √13 | e ₂ > + 2 / √13 | e ₃ >) + | e ₁ >
= (√2 + 1) | e ₁ > + (√2 + 6 / √13) | e ₂ > + 4 / √13 | e ₃ >
. . . (Formula 11)
It becomes. Here, since n ₃ is a unit vector corresponding to the general noun “music” and the basic attribute element corresponding to the general noun “music” is “music”, n ₃ is equal to | e ₁ >.

As a result, the percentage of content attributes “music”, “computer”, and “linguistics”
√2 + 1: (√2 + 6 / √13): 4 / √13
≒ 37%: 47%: 16%. . . (Formula 12)
It becomes.

  Therefore, when it is necessary to determine one attribute related to the content, “computer” is selected. When two attributes related to the content are determined, “music” and “computer” may be selected.

  The functional configuration of the attribute evaluation system according to the second modification is almost the same as the functional configuration shown in FIG. However, in the second modification, the vector information generation unit 602 extracts a general noun by performing morphological analysis on the text and image metadata included in the content, and the appearance frequency of the extracted general noun Is stored in the storage unit 601.

  Then, as described in FIG. 1, the vector information generation unit 602 performs processing for expressing the extracted general nouns using base vectors that are orthogonal to each other and calculating a position in the vector space.

  Further, the attribute evaluation unit 603 uses the expression 8 to calculate the content attributes and the general nouns from the appearance frequency of the general nouns and the position information in the vector space of the general nouns expressed using mutually orthogonal basis vectors. Evaluate the attributes of users accessing the content.

  Further, the processing procedure of the attribute evaluation processing according to the second modification is almost the same as the processing procedure shown in FIG. However, in the second modification, in step S103, the vector information generation unit 602 extracts general nouns from the text and image metadata included in the content, and the frequency of appearance of the extracted general nouns in the storage unit 601. Perform the process of memorizing.

  In step S106, the attribute evaluation unit 603 calculates the sum of the vectors corresponding to each general noun using Expression 8 instead of Expression 5. Further, in step S107, the attribute evaluation unit 603 evaluates the attribute related to the content including the text and the image from the coefficient of each base vector in the vector sum as described in Expression 10.

  As described above, in the second modification, the attribute evaluation unit 603 evaluates attributes related to content based on general nouns extracted from text and image metadata included in the content. By further using the metadata of the image, it is possible to evaluate the attribute relating to the content more accurately.

(Modification 3 of an Example)
By the way, in the said Example and the modification of an Example, although it was decided to extract a general noun from the content or content metadata which evaluates an attribute, and evaluated an attribute based on the extracted general noun, the 1st content It is also possible to extract general nouns from the second content associated with the hyperlink or trackback and evaluate the attributes of the first content.

  Specifically, the attribute evaluation process as described in FIG. 1 is performed on the second content, the attribute related to the second content is evaluated, and the first content is determined based on the evaluation result. Evaluate attributes. Here, the second content is one or a plurality of contents.

  For example, when the attributes related to the second plurality of contents are evaluated as “music”, “computer”, and “linguistics”, and the most content with the attribute “computer” is evaluated, the first The attribute relating to the content may be evaluated as “computer”.

  The functional configuration of the attribute evaluation system according to Modification 3 is substantially the same as the functional configuration shown in FIG. However, in the third modification, the vector information generation unit 602 performs morphological analysis on the second content, extracts general nouns from the second content, and stores the appearance frequency of the extracted general nouns. Store in the unit 601.

  Then, as described in FIG. 1, the vector information generation unit 602 performs processing for expressing the extracted general nouns using base vectors that are orthogonal to each other and calculating a position in the vector space.

  In addition, the attribute evaluation unit 603 uses the expression 8 to calculate the second content from the appearance frequency of the general noun and the position information in the vector space of the general noun expressed using mutually orthogonal basis vectors. The attribute related to the first content is evaluated based on the attribute related to the second content.

  Further, the processing procedure of the attribute evaluation processing according to the third modification is substantially the same as the processing procedure shown in FIG. However, in the third modification, the processing from step S101 to step S108 is executed for the second content, and then the first content is related based on the attribute information evaluated for the second content. Evaluate attributes.

  As described above, in the third modification of the present embodiment, the vector information generation unit 602 sets the attribute related to the second content associated with the first content whose attribute is evaluated by the hyperlink or the trackback. Information relating to a vector representing a position in the vector space of the general noun related to the second content based on mutually orthogonal vectors associated with each candidate is generated, and the attribute evaluation unit 603 generates vector information Since the attribute related to the first content is evaluated based on the information generated by the generation unit 602, the attribute of the first content is flexibly and efficiently determined from the second content associated by the hyperlink or the trackback. Therefore, it is possible to accurately and dynamically evaluate the attribute relating to the content.

  As described above, the case where the attribute evaluation process is realized on the computer has been described. However, the program for realizing the attribute evaluation process is recorded on a computer-readable recording medium, and the program recorded on the recording medium is stored in the computer. The attribute evaluation process may be realized by reading and executing. FIG. 8 is a block diagram illustrating a hardware configuration of the computer 100 that realizes attribute evaluation processing.

  As shown in FIG. 8, the computer 100 includes a CPU 110 that executes the program, an input device 120 that inputs data, a ROM 130 that stores various data, a RAM 140 that stores calculation parameters and the like, and attribute evaluation processing. A reading device 150 that reads a program from a recording medium 200 that records a program for realizing, an output device 160 such as a display, and a network interface 170 that exchanges data with other computers via a network 300. The bus 180 is connected.

  The CPU 110 implements attribute evaluation processing by reading a program recorded on the recording medium 200 via the reading device 150 and then executing the program. Examples of the recording medium 200 include an optical disk, a flexible disk, a CD-ROM, and a hard disk. Further, this program may be installed in the computer 100 via the network 300.

  Although the embodiments of the present invention have been described so far, the present invention may be implemented in various different embodiments in addition to the above-described embodiments within the scope of the technical idea described in the claims. It ’s good.

  In addition, among the processes described in this embodiment, all or part of the processes described as being performed automatically can be performed manually, or the processes described as being performed manually can be performed. All or a part can be automatically performed by a known method.

  In addition, the processing procedure, control procedure, specific name, and information including various data and parameters shown in the above-mentioned document and drawings can be arbitrarily changed unless otherwise specified.

  Each component of each illustrated device is functionally conceptual and does not necessarily need to be physically configured as illustrated. In other words, the specific form of distribution / integration of each device is not limited to that shown in the figure, and all or a part thereof may be functionally or physically distributed or arbitrarily distributed in arbitrary units according to various loads or usage conditions. Can be integrated and configured.

  Furthermore, all or some of the processing functions performed in each device may be realized by a CPU and a program that is analyzed and executed by the CPU, or may be realized as hardware by wired logic.

  As described above, the attribute evaluation apparatus, the attribute evaluation method, and the attribute evaluation program according to the present invention are attribute evaluations that require flexible, efficient, accurate, and dynamic evaluation of content attributes. Useful for systems.

It is a figure explaining the concept of the attribute evaluation process concerning this invention. It is a figure which shows an example of the execution result of a morphological analysis. It is a figure which shows an example of the digital thesaurus which memorize | stored the information of the ratio between basic attribute elements and basic attribute elements. It is a figure explaining the attribute evaluation method which concerns on the content at the time of using the base vector which is not mutually orthogonal. It is a figure which shows the function structure of the attribute evaluation system 20 which concerns on a present Example. It is a flowchart which shows the process sequence of the attribute evaluation process which concerns on a present Example. It is a figure which shows an example of the execution result of the morphological analysis in the modification 1. It is a block diagram which shows the hardware constitutions of the computer 100 which implement | achieves attribute evaluation processing.

Explanation of symbols

10a to 10c External web server 20 Attribute evaluation system 30 Internet 40a to 40c Web server 50 Information collection server 500 Data transmission / reception unit 501 Information collection unit 502 Storage unit 502a Link information 502b Metadata information 502c Content information 503 Control unit 60 Attribute evaluation server 600 Data transmission / reception unit 601 Storage unit 601a Digital thesaurus data 601b Attribute evaluation information 602 Vector information generation unit 603 Attribute evaluation unit 604 Control unit 70 Advertising server 700 Data transmission / reception unit 701 Storage unit 701a Advertising data 701b Advertising distribution data 702 Advertising distribution processing unit 703 Control Part 80 LAN

Claims (30)

An attribute evaluation apparatus that evaluates attributes related to content based on information related to content,
Vector information generating means for generating information related to a vector representing a position in a vector space of information related to the content based on mutually orthogonal vectors associated with each candidate of the attribute related to the content;
Attribute evaluation means for evaluating an attribute relating to the content based on information generated by the vector information generation means;
An attribute evaluation apparatus characterized by comprising:   The database further includes a database that stores information related to content and information about candidate attributes related to the content in association with each other, and the vector information generation unit uses the information related to the content as a search key to store information about candidate attributes related to the content. The attribute evaluation apparatus according to claim 1, wherein information relating to the vector is generated based on information on candidate attributes relating to the content obtained as a result of retrieval from the database.   The database stores a plurality of different combinations between information related to content and attribute candidates related to the content, and the vector information generation unit is configured to select candidate attributes related to the content in a specified combination among the plurality of combinations. The attribute evaluation apparatus according to claim 2, wherein information is searched, and information related to the vector is generated based on information on candidate attributes related to the content obtained as a result of the search.   The database further stores information relating to the weight of each candidate for the attribute relating to the content, and the vector information generating means reads out the information relating to the candidate for the attribute relating to the content and the information relating to the weight from the database, and reads the information. The attribute evaluation apparatus according to claim 2, wherein information related to the vector is generated based on the received information.   5. The attribute evaluation device according to claim 1, wherein the attribute evaluation unit evaluates a plurality of attributes related to the content and a priority of each attribute with respect to the predetermined content. .   The attribute evaluation unit sets a weight of information related to the vector based on an appearance frequency of information related to the content, and evaluates an attribute related to the content based on the set weight. The attribute evaluation apparatus as described in any one of 1-5.   The vector information generation unit generates information related to the vector in a vector space whose dimension increases in accordance with the increase in the number of candidates when the number of attribute candidates related to the content increases. The attribute evaluation apparatus as described in any one of Claims 1-6.   The attribute evaluation apparatus according to claim 1, wherein the information relating to the content is metadata including information relating to the content of the content or information extracted from the content.   The vector information generation means is based on mutually orthogonal vectors associated with respective attribute candidates related to the second content associated with the first content for which the attribute is evaluated by hyperlink or trackback, And the information concerning the vector showing the position in the vector space of the information which concerns on the said 2nd content is produced | generated, The said attribute evaluation means is a said 1st based on the information produced | generated by the said vector information production | generation means The attribute evaluation apparatus according to claim 1, wherein an attribute related to content is evaluated.   The attribute evaluation apparatus according to claim 1, wherein the information related to the content is a search word used for searching the content or a word for which a hyperlink is set. . An attribute evaluation method for evaluating attributes related to content based on information related to content,
A vector information generating step for generating information related to a vector representing a position in a vector space of information related to the content based on mutually orthogonal vectors associated with each candidate for the attribute related to the content;
An attribute evaluation step for evaluating an attribute relating to the content based on the information generated by the vector information generation step;
Attribute evaluation method characterized by including.   The vector information generation step searches for information on candidate attributes related to content from a database that stores information related to the content and information on candidate attributes related to the content in association with the information related to the content as a search key, The attribute evaluation method according to claim 11, wherein the information related to the vector is generated based on information on candidate attributes related to the content obtained as a result of the search.   The vector information generation step searches for information on attribute candidates related to the content in the specified combination from a database storing a plurality of different combinations between the information related to the content and the attribute candidates related to the content. 13. The attribute evaluation method according to claim 12, wherein information related to the vector is generated on the basis of information on candidate attributes related to the content obtained as a result.   The vector information generation step reads out and reads out the information on the candidate attributes and the information on the weights from the database storing the information on the candidate attributes on the contents and the information on the weights of the candidate attributes on the contents. The attribute evaluation method according to claim 12 or 13, wherein information on the vector is generated based on information.   The attribute evaluation method according to claim 11, wherein the attribute evaluation step evaluates a plurality of attributes related to the content with respect to predetermined content and a priority of each attribute. .   The attribute evaluation step sets the weight of the information related to the vector based on the appearance frequency of the information related to the content, and evaluates the attribute related to the content based on the set weight. The attribute evaluation method as described in any one of 11-15.   In the vector information generation step, when the number of attribute candidates related to content increases, information related to the vector in a vector space whose dimension increases in accordance with the increase in the number of candidates is generated. The attribute evaluation method according to any one of claims 11 to 16.   The attribute evaluation method according to claim 11, wherein the information relating to the content is metadata including information relating to the content of the content or information extracted from the content.   The vector information generation step is based on mutually orthogonal vectors associated with each candidate for the attribute related to the second content associated with the first content whose attribute is evaluated by hyperlink or trackback, And the information which concerns on the vector showing the position in the vector space of the information which concerns on the said 2nd content is produced | generated, The said attribute evaluation process is a said 1st based on the information produced | generated by the said vector information production | generation process The attribute evaluation method according to claim 11, wherein an attribute related to the content is evaluated.   The attribute evaluation method according to claim 11, wherein the information related to the content is a search word used for searching the content or a word for which a hyperlink is set. . An attribute evaluation program for evaluating attributes related to content based on information related to content,
A vector information generation procedure for generating information related to a vector representing a position in a vector space of information related to the content based on mutually orthogonal vectors associated with each candidate for the attribute related to the content;
An attribute evaluation procedure for evaluating an attribute related to the content based on information generated by the vector information generation procedure;
An attribute evaluation program for causing a computer to execute.   The vector information generation procedure searches for information on attribute candidates related to the content from a database that stores information related to the content and information about candidate attributes related to the content in association with the information related to the content as a search key, The attribute evaluation program according to claim 21, wherein information related to the vector is generated based on attribute candidate information related to the content obtained as a result of the search.   The vector information generation procedure searches for information on attribute candidates related to content in a specified combination from a database that stores a plurality of different combinations between information related to content and attribute candidates related to the content. 23. The attribute evaluation program according to claim 22, wherein information related to the vector is generated on the basis of information on candidate attributes related to the content obtained as a result.   The vector information generation procedure reads the information on the candidate attributes and the information on the weights from the database storing the information on the candidate attributes on the contents and the information on the weights of the candidate candidates on the contents, and reads the information The attribute evaluation program according to claim 22 or 23, wherein information concerning said vector is generated based on information.   The attribute evaluation program according to any one of claims 21 to 24, wherein the attribute evaluation procedure evaluates a plurality of attributes related to content with respect to predetermined content and a priority of each attribute. .   The attribute evaluation procedure sets a weight of information related to the vector based on an appearance frequency of information related to the content, and evaluates an attribute related to the content based on the set weight. The attribute evaluation program as described in any one of 21-25.   In the vector information generation procedure, when the number of attribute candidates related to content increases, information related to the vector in a vector space whose dimensions increase in accordance with the increase in the number of candidates is generated. The attribute evaluation program according to any one of claims 21 to 26.   The attribute evaluation program according to any one of claims 21 to 27, wherein the information relating to the content is metadata including information relating to the content of the content or information extracted from the content.   The vector information generation procedure is based on vectors orthogonal to each other associated with each candidate for the attribute relating to the second content associated with the first content for which the attribute is evaluated by hyperlink or trackback, And the information concerning the vector showing the position in the vector space of the information concerning the second content is generated, and the attribute evaluation procedure is based on the information generated by the vector information generation procedure. The attribute evaluation program according to any one of claims 21 to 28, wherein an attribute relating to the content is evaluated.   The attribute evaluation program according to any one of claims 21 to 28, wherein the information related to the content is a search word used for searching the content or a word for which a hyperlink is set. .

Images