JP2008123043A - Information processing apparatus and method, program, and recording medium - Google Patents

Abstract

To easily and more accurately classify attribute values of attributes assigned to content.
A performer feature quantity vector creating unit creates a feature quantity indicating a feature of an attribute value of an attribute given to content from information about the content. The classification unit 72-1 selects a target attribute value that is an attribute value to be classified as an attribute value that has the same attribute as the attribute of the target attribute value, and is a pre-classified attribute value. Of these, classification is performed according to the classification of the attribute values that have been classified into the feature amounts similar to the feature amounts of the target attribute value. The classification result correction unit 91 corrects the classification of the target attribute value according to the classification bias of the classified attribute value assigned to the content that satisfies the predetermined condition. The present invention can be applied to a server.
[Selection] Figure 10

Description

  The present invention relates to an information processing apparatus and method, a program, and a recording medium, and more particularly, to an information processing apparatus and method, a program, and a recording medium that can classify performers who appear in content.

  Traditionally, in a system that recommends programs to be broadcast, it has appeared in comedy systems that appear in comedy programs, idols that sing so-called popular songs and gravure, moderators that host programs, and love dramas A type such as romance is manually given to the performer, and a broadcast program is recommended based on the type given to the performer.

  Of the performers who appear in the program that the user has watched, use the viewing history of the program in which the performer whose type has been manually assigned is used to display the type of person often seen for the reason for recommending the program Some have done so (for example, see Patent Document 1)

JP-A-2005-56361

  When a type is manually assigned to a performer, for a performer who is not assigned a type, even if the user is watching a program in which the performer appears, the type is assigned to the performer. Therefore, the type of the performer could not be reflected in the type of performer that the user often sees.

  Therefore, if it is attempted to manually give a type to all the performers, it will take enormous effort. Further, when a type is manually assigned to a large number of performers, it is difficult to assign a type to performers based on a unified standard, and it is difficult to accurately classify performers. Each performer is indicated by an attribute value of one attribute assigned to the content.

  The present invention has been made in view of such a situation, and makes it possible to classify attribute values of attributes assigned to content easily and more accurately.

  An information processing apparatus according to an aspect of the present invention includes a feature amount creating unit that creates a feature amount indicating a feature of an attribute value of an attribute assigned to the content from information about the content, and an attribute value to be classified. A certain target attribute value is an attribute value of the same attribute as the attribute of the target attribute value, and is similar to the feature amount of the target attribute value among classified attribute values that are pre-classified attribute values A first classification unit configured to classify according to a classification of the classified attribute value of the feature quantity; and the target attribute according to a classification bias of the classified attribute value assigned to the content satisfying a predetermined condition Correction means for correcting the classification of values.

  The correction means may correct the classification of the target attribute value according to the classification bias of the classified attribute value of the attribute assigned to the content satisfying the condition among all the contents. it can.

  In the correction means, in the bias of the classification of the classified attribute value of the attribute that is given to the content that satisfies the condition that satisfies the condition that limits the attribute of the content in which the predetermined performer indicated by the target attribute value appears Accordingly, the classification of the target attribute value can be corrected.

  In the correction means, the classification of the target attribute value indicating a predetermined performer who appears in the content that satisfies the condition satisfies the classification indicating a performer that appears in the content that satisfies the condition. The classification of the target attribute value can be corrected so as to be closer to the bias of the classification of the completed attribute value.

  The correction means includes a bias larger than the first threshold when the bias of classification of the classified attribute value indicating a performer appearing in the content satisfying the condition is larger than a predetermined first threshold. The classification of the target attribute value can be corrected according to the classification of the classified attribute value.

  In the correction means, the bias of classification of the classified attribute value indicating a performer appearing in the content satisfying the condition is larger than a predetermined first threshold value, and the content satisfying the condition is satisfied. A difference between a bias in classification of the classified attribute value indicating a performer and a bias in classification of the target attribute value indicating a performer in the content satisfying the condition is determined in advance. If the threshold value is larger than the first threshold value, the classification of the target attribute value can be corrected according to the classification of the classified attribute value biased larger than the first threshold value.

  The first classification means includes the classified attribute of the feature quantity that is determined to be similar to the feature quantity of the target attribute value based on a similarity weighted for each element of the feature quantity including a plurality of elements. The target attribute value can be classified according to the value classification.

  The first classifying unit classifies the feature value similar to the feature value of the target attribute value indicating a predetermined performer among a plurality of types into the first type of the classified attribute value. Accordingly, the target attribute value indicating the performer can be classified into the first type.

  The classification attribute value of the feature quantity similar to the feature quantity of the target attribute value indicating the performer among the plurality of types by a scheme different from the classification scheme of the first classification means. A second classification unit that classifies the target attribute value indicating the performer into the second type according to classification into two types, and either the first classification unit or the second classification unit It is possible to further provide selection means for selecting one of the types for each type.

  The feature quantity creating means creates the feature quantity indicating a feature about the attribute value indicating a performer who appears in the content that is a program from information about the content that is a broadcast program, According to the bias of the classification of the classified attribute value indicating the performer who is appearing in the content that is a program broadcast from a predetermined broadcasting station in a predetermined time zone, the correction means Classification can be corrected.

  According to an information processing method of one aspect of the present invention, a feature amount indicating a feature of an attribute value of an attribute given to the content is created from information about the content, and a target attribute value that is an attribute value to be classified is obtained. The classification of the feature quantity that is similar to the feature quantity of the target attribute value among the sorted attribute values that are the attribute values of the same attribute as the attribute of the target attribute value and are pre-classified attribute values And classifying the target attribute value according to the classification bias of the classified attribute value assigned to the content satisfying a predetermined condition.

  The program according to one aspect of the present invention creates a feature amount indicating a feature of an attribute value of an attribute given to the content from information about the content, and sets the target attribute value that is an attribute value to be classified as the target attribute value. Of the attribute values of the same attribute as the attribute of the target attribute value, the classified attribute of the feature quantity that is similar to the feature quantity of the target attribute value among the sorted attribute values that are the attribute values that are classified in advance Classifying according to the value classification, and causing the computer to execute a step of correcting the classification of the target attribute value according to the classification bias of the classified attribute value given to the content satisfying a predetermined condition.

  A recording medium program according to an aspect of the present invention creates a feature amount indicating a feature of an attribute value of an attribute assigned to the content from information about the content, and a target attribute value that is an attribute value to be classified The attribute value of the same attribute as the attribute of the target attribute value, and among the classified attribute values that are pre-classified attribute values, the feature quantity similar to the feature quantity of the target attribute value Information including a step of classifying according to the classification of the classified attribute value and correcting the classification of the target attribute value according to the classification bias of the classified attribute value given to the content satisfying a predetermined condition Have the computer execute the process.

  In one aspect of the present invention, a feature amount indicating a feature of an attribute value of an attribute assigned to the content is created from information about the content, and the target attribute value that is an attribute value to be classified is the target The attribute value having the same attribute as the attribute value attribute, and the classified attribute value of the feature quantity that is similar to the feature quantity of the target attribute value among the classified attribute values that are attribute values that have been classified in advance The classification of the target attribute value is corrected according to the classification bias of the classified attribute value assigned to the content satisfying a predetermined condition.

  As described above, according to one aspect of the present invention, performers can be classified.

  Further, according to one aspect of the present invention, it is possible to easily and more accurately classify attribute values of attributes assigned to content.

  Embodiments of the present invention will be described below. Correspondences between constituent elements of the present invention and the embodiments described in the specification or the drawings are exemplified as follows. This description is intended to confirm that the embodiments supporting the present invention are described in the specification or the drawings. Therefore, even if there is an embodiment which is described in the specification or the drawings but is not described here as an embodiment corresponding to the constituent elements of the present invention, that is not the case. It does not mean that the form does not correspond to the constituent requirements. Conversely, even if an embodiment is described here as corresponding to a configuration requirement, that means that the embodiment does not correspond to a configuration requirement other than the configuration requirement. It's not something to do.

  The information processing apparatus according to one aspect of the present invention is a feature amount creating unit that creates a feature amount indicating a feature of an attribute value of an attribute given to the content from information about the content (for example, a performer feature amount in FIG. 10). A vector creation unit 32), and a target attribute value that is an attribute value to be classified, an attribute value of the same attribute as the attribute of the target attribute value, and a classified attribute value that is a pre-classified attribute value First classifying means (for example, the classifying unit 72-1 of FIG. 10) for classifying according to the classification of the classified attribute value of the feature quantity similar to the feature quantity of the target attribute value, and a predetermined Correction means (for example, the classification result correction unit 91 in FIG. 10) that corrects the classification of the target attribute value according to the classification bias of the classified attribute value assigned to the content that satisfies the condition .

  The classification attribute value of the feature quantity similar to the feature quantity of the target attribute value indicating the performer among the plurality of types by a scheme different from the classification scheme of the first classification means. A second classification unit (for example, the classification unit 72-N in FIG. 10) that classifies the target attribute value indicating the performer into the second type according to the classification into the second type, and the first The selection means (for example, the selection part 71 of FIG. 10) which selects any one of the said classification means or said 2nd classification means for every type can be further provided.

  An information processing method and program according to one aspect of the present invention creates a feature amount indicating a feature of an attribute value of an attribute assigned to the content from information about the content (for example, step S16 in FIG. 2), A target attribute value that is a target attribute value is an attribute value of the same attribute as the attribute of the target attribute value, and among the classified attribute values that are pre-classified attribute values, Classification according to the classification of the attribute values that have been classified similar to the feature values (for example, step S37 in FIG. 5), and classification of the classified attribute values that are given to the content that satisfies a predetermined condition The classification of the target attribute value is corrected according to the bias (for example, step S56 in FIG. 8).

  FIG. 1 is a block diagram illustrating an example of a configuration of an information processing apparatus 11 according to an embodiment of this invention.

  The information processing apparatus 11 classifies attribute values of attributes assigned to the content.

  Here, the contents include not only moving images or sounds such as programs broadcast from television broadcast stations, programs broadcast from radio broadcast stations, movies, or music songs, but also books or still images. Information that is significant for the person is sufficient.

  The attribute assigned to the content refers to an item indicating an artist, a performer, a lyricist, a composer, a director, a screenwriter, a director, a label, a broadcasting station, or a production. The attribute value of the attribute assigned to the content includes a predetermined artist, a predetermined performer, a predetermined lyricist, a predetermined composer, a predetermined director, a predetermined screenwriter, a predetermined director, a predetermined label, a predetermined label, Indicates a broadcasting station or a predetermined production. That is, for example, the attribute value of the attribute indicating the artist among the attributes assigned to the content indicates a specific artist among the artists.

  In other words, for example, the information processing apparatus 11 classifies attribute values of attributes assigned to the content, which indicate persons or groups involved in the creation of the content (a group or organization that is not a natural person, a corporation, or a legal person). .

  More specifically, for example, by the information processing apparatus 11, for a content that is a program, a predetermined performer, a predetermined producer, a predetermined director, a predetermined screenwriter, a predetermined original author, a predetermined sponsor, a predetermined production Or a predetermined broadcast station or the like is classified. Further, the information processing apparatus 11 classifies a predetermined writer, a predetermined publisher, a predetermined editor, or the like for the content that is a book. Further, the information processing apparatus 11 uses a predetermined performer such as a performer, a singer, or an artist (including a group), a predetermined producer, a predetermined composer, a predetermined songwriter, or a predetermined label for content that is music. Etc. are classified. Furthermore, the information processing apparatus 11 classifies a specific director, a specific original author, a specific production company, a specific distribution company, or the like for content that is a movie.

  Hereinafter, as content, a program broadcast from a broadcasting station which is a television broadcasting station will be described as an example, and as an attribute value of an attribute assigned to the content, an attribute value indicating a predetermined performer will be described as an example.

  The information processing apparatus 11 includes a content information database 31, a performer feature vector generation unit 32, a classification unit 33, and a performer type database 34.

  The content information database 31 holds information about content (hereinafter referred to as content information). For example, the content information database 31 includes the genre, the broadcast start time and the broadcast end time and broadcast time, the broadcast station, performer, title, or summary of the content that is the broadcast program. It holds content information consisting of the attribute values shown. For example, the content information database 31 holds content information that is metadata of a program.

  In addition, the attributes of the content information are the performance relationship between the performers, the appearance genre, the appearance program, the role in the appearance program such as the moderator, narration and reporter, the appearance program information including the summary and commentary of the program, etc. It may be a profile of a performer including gender and affiliated office.

  Importance is given to each attribute value of the content information held in the content information database 31. For example, content information is analyzed in advance for content that is a program in which all performers appear for a certain period of time, such as half a year or one year, and the attribute values such as title, genre, and performer are set to TF-IDF (term frequency -inverse document frequency) and so on.

  The performer feature quantity vector creation unit 32 creates a feature quantity indicating a feature about an attribute value of an attribute given to the content from the information about the content. For example, the performer feature value vector creation unit 32 acquires content information about content from the content information database 31 and creates the feature amount of the performer. For example, the feature amount of a performer is represented by a vector (hereinafter referred to as a performer feature amount vector).

  That is, the performer feature value vector creation unit 32 creates a feature value indicating the feature of the performer who appears in the content from the content information about the content. For example, the performer feature value vector creation unit 32 acquires performer information and co-starring information, which is information indicating the performer, from the content information about the content, and creates the feature amount of the performer.

  The classification unit 33 selects a target attribute value that is an attribute value to be classified as an attribute value of the same attribute as the attribute of the target attribute value, and among the classified attribute values that are pre-classified attribute values. Then, the classification is performed according to the classification of the classified attribute value of the feature quantity similar to the feature quantity of the target attribute value.

  For example, the classification unit 33 classifies performers. More specifically, the classification unit 33 may include a comedy system appearing in a comedy program, an idol system appearing in a song or gravure, a moderator system hosting a program, or a romantic system appearing in a love drama. Classify performers into types.

  More specifically, the classification unit 33 reads out information indicating the classification of performers classified in advance from the performer type database 34, and based on the read information indicating the classification of performers classified in advance. Categorize other performers. That is, the classification unit 33 selects a target performer that is a subject to be classified as a feature amount similar to the feature amount of the target performer among the classified performers that are pre-classified performers. Categorize according to the categorized performers. The classification unit 33 supplies the performer classification result to the performer type database 34 or outputs it to the outside.

  The performer type database 34 holds the result of performer classification. In addition, the performer type database 34 holds information indicating the classification of performers classified manually by a predetermined type as learning data.

  First, a process for creating a performer feature vector from content information will be described.

  FIG. 2 is a flowchart for explaining a process for creating a performer feature quantity vector. In step S11, the performer feature vector generation unit 32 acquires a performer list that is a list of all performers.

  In step S11, the performer feature vector generation unit 32 acquires content information about the content from the content information database 31, and acquires performer information or co-star information from the content information, thereby listing all performers. May be generated.

  In step S12, the performer feature vector generation unit 32 determines whether or not each performer feature vector of all performers in the performer list has been created, and each performer in the performer list is determined. If it is determined that no performer feature vector has been created, the procedure proceeds to step S13. In step S <b> 13, the performer feature vector generation unit 32 selects one performer who has not yet created a performer feature vector from the performers in the performer list.

  In step S14, the performer feature vector generation unit 32 acquires a program list that is a list of programs in which the selected performer has appeared.

  In step S14, the performer feature vector generation unit 32 acquires content information about the content from the content information database 31, and collates the performer information or co-starring information included in the content information with the selected performer. Thus, a program in which the selected performer has appeared may be specified, and a program list that is a list of all programs in which the selected performer has appeared may be generated.

  In step S <b> 15, the performer feature vector generation unit 32 acquires content information of all programs in the program list from the content information database 31.

  FIG. 3 is a diagram illustrating an example of content information. For example, content information includes a content ID for specifying content, an attribute ID for specifying an attribute, an attribute value indicating the value of the content attribute, an update count indicating the number of times the attribute value has been updated, and an attribute value It consists of importance. In the content information shown in FIG. 3, examples of one attribute of the content information shown in the upper one row in FIG. 3 are a content ID of 2000114789142, an attribute ID of 1, an attribute value of 153144, and 1 It consists of a certain number of updates and an importance level of 1.58454294474. That is, the attribute value of the attribute specified by the attribute ID of 1 of the content specified by the content ID of 2000114789142 is 153144, the attribute value is updated once, and the importance of the attribute value is 1.58454294474 is shown by the upper horizontal row in the content information of FIG.

  In step S <b> 16, the performer feature vector generation unit 32 adds the content information of all the acquired programs by weighting the attribute value with a predetermined weight for each attribute. In this case, for example, the weight for weighting the attribute value is different for each attribute. By weighting, the effect of the attribute can be changed for each attribute. And the performer feature-value vector preparation part 32 adds together the content information which consists of a some attribute value about each of all the programs in which the selected performer appears for every attribute, and for every attribute. , And obtain a result consisting of the sum of the respective attribute values.

  That is, in step S <b> 16, the performer feature quantity vector creation unit 32 creates a feature quantity indicating a feature about the attribute value of the attribute given to the content from the information about the content.

  In step S17, the performer feature vector generation unit 32 stores the result of adding in step S16 in the performer type database 34 as the performer feature vector of the selected performer. That is, the performer feature vector generation unit 32 performs, for each attribute, the result of the sum of the attribute values of the content information as the performer feature vector having the sum of the attribute values as an element. 34.

  After step S17, the procedure returns to step S12.

  If it is determined in step S12 that the performer feature quantity vectors of all the performers in the performer list have been created, the process of creating the performer feature quantity vectors ends.

  FIG. 4 is a diagram illustrating an example of performer feature quantity vectors stored in the performer type database 34. For example, the performer feature vector is represented by a performer ID for identifying a performer, an attribute ID for identifying an attribute, an attribute value indicating the attribute value of the performer, and the importance of the attribute value. .

  In the performer feature vector shown in FIG. 4, examples of one attribute of the content information shown in the upper horizontal row of FIG. 4 are performer ID 3, attribute ID 4, and attribute 1 Consists of a value and an importance level of 6.822980305192. That is, the attribute value of the attribute identified by the attribute ID of 4 of the performer identified by the performer ID of 3 is 1, and the importance of the attribute value is 6.822980305192. This is indicated by the upper horizontal row of the four performer feature vectors.

  Next, the process of classifying performers by the classifying unit 33 will be described with reference to the flowchart of FIG. In step S31, the classification unit 33 determines whether or not a type vector has been assigned to all performers. If it is determined that no type vector has been assigned to all performers, the performer is not classified. Therefore, the procedure proceeds to step S32.

  Here, the type vector is information indicating the classification of performers, and includes one or a plurality of elements, each of which indicates the type of performer. For example, a type vector is an element that indicates whether or not a comedy system appears in a comedy program, an element that indicates whether or not an idol system appears on a singing or gravure, and a moderator system that moderates the program It is composed of an element indicating whether or not and an element indicating whether or not it is a romance system appearing in a love drama.

  Each element of the type vector is set to 1 indicating that it is the type or 0 indicating that it is not the type.

  In addition, each element of the type vector may be any value in the range of 0 to 1 and a value indicating the degree or probability of the type.

  In step S32, the classification unit 33 selects performers to which the type vector is assigned, that is, performers to be classified from performers to which the type vector is not yet assigned (performers not classified).

  In step S <b> 33, the classification unit 33 acquires the performer feature value vector of the target performer who is the selected performer from the performer type database 34.

  In step S <b> 34, the classification unit 33 determines the similarity between each performer feature value vector of each classified performer who is a performer to which a type vector has been assigned in advance, and the performer feature value vector of the target performer. calculate.

  Here, the classified performers are manually classified into predetermined types, and a type vector indicating the classified type is stored in advance in the performer type database 34 as learning data.

  That is, in step S <b> 34, the classification unit 33 reads each performer feature value vector of the classified performers from the performer type database 34. And the classification | category part 33 calculates the similarity degree of each performer feature-value vector of the classified performer read from the performer type database 34, and the performer feature-value vector of an object performer.

For example, the classification unit 33 obtains a cosine measure as the similarity sim (X, Y) by the calculation represented by the equation (1).

In Expression (1), X represents a performer feature value vector of the target performer, and Y represents a performer feature value vector of the classified performer. Further, a indicates an attribute of the performer feature value vector, A indicates a set of attributes a, and X _a indicates an attribute value of the attribute a of the performer feature value vector X. Similarly, Y _a indicates the attribute value of the attribute a of the performer feature value vector Y. Further, W _a indicates the weight of the attribute a.

  In step S35, the classification unit 33 selects K performers (K is a positive integer) in order of similarity from the similarity calculation results in step S34.

  In step S <b> 36, the classification unit 33 acquires the type vector of the selected K performers that have been classified from the performer type database 34.

  In step S <b> 37, the classification unit 33 calculates the type vector of the target performer from the type vectors of the selected K performers.

式（２）において、iは、タイプベクトルの要素を示し、f_x(C_i)は、所定の閾値t_iを超える場合、1を出力し、閾値t_i以下である場合、0を出力する関数を示す。また、f_k(C_i)は、所定の閾値t_iを超える場合、1を出力し、閾値t_i以下である場合、0を出力する関数を示す。閾値t_iは、i毎に異なるか、または同じ値のi毎の値とされる。 For example, the classification unit 33 calculates the type vector of the target performer by the K-nearest Neighbor method. Expression (2) is an expression showing the calculation of the type vector of the target performer by the K-nearest Neighbor method. Equation (2) represents the calculation of one element of the type vector elements.

In Expression (2), i represents an element of the type vector, and f _x (C _i ) outputs 1 when it exceeds a predetermined threshold value t _i , and outputs 0 when it is equal to or less than the threshold value t _i. Indicates a function. Further, f _k (C _i ) represents a function that outputs 1 when it exceeds a predetermined threshold value t _i and outputs 0 when it is equal to or less than the threshold value t _i . The threshold value t _i is different for each i or the same value for each i.

In equation (2), f _k (C _i ) indicates a value that is either 0 or 1 of the element i of the type vector of the kth classified performer among the K classified performers. . That is, the average value of the element i of the type vector of the selected K performers is set as the value of the element i of the type vector of the target performer.

  In this way, in step S37, the classification unit 33 sets the target attribute value, which is the attribute value to be classified, as the attribute value having the same attribute as the attribute of the target attribute value, The classification is performed according to the classification of the classified attribute value of the feature quantity similar to the feature quantity of the target attribute value among the classified attribute values.

  In step S <b> 38, the classification unit 33 assigns the type vector calculated in step S <b> 37 to the target performer and stores it in the performer type database 34.

  After step S38, the procedure returns to step S31.

  If it is determined in step S31 that the type vector has been assigned to all performers, the performer classification process ends.

  In this way, performers can be classified.

式（３）において、iは、タイプベクトルの要素を示し、P(x_j｜C_i)は、タイプベクトルで示される分類C_iが起きた場合の属性値X_jの起こる確率を示す。 Note that, for example, the classification unit 33 may calculate the type vector of the target performer according to Equation (3) as a naive Bayes classifier.

In Expression (3), i represents an element of the type vector, and P (x _j | C _i ) represents the probability of occurrence of the attribute value X _j when the classification C _i indicated by the type vector occurs.

That is, if the weight of the type vector of a performer who has already been classified is considered as the probability of occurrence of the attribute value X _j when the classification C _i occurs, the type vector of the target performer can be calculated as a posterior probability from equation (3).

  Further, the classification unit 33 may classify the target performer by calculating the type vector not only by the K-nearest Neighbor method or the Naive Bayes method but by other methods such as a support vector machine. The classification of target performers is a so-called classification problem, and methods for classifying blog reputation information or images using text on the Web can be applied to target performer classification. .

  Furthermore, a classification unit may be provided for each type vector element.

  FIG. 6 is a block diagram illustrating another example of the configuration of the information processing apparatus 11 when a classification unit is provided for each element of a type vector. 6, the same parts as those shown in FIG. 1 are denoted by the same reference numerals, and the description thereof is omitted.

  When a classification unit is provided for each type vector element, the information processing apparatus 11 includes a content information database 31, a performer feature value vector creation unit 32, a performer type database 34, and a selection classification unit 51.

  The selection classification unit 51 classifies performers based on the type vector. The selection classification unit 51 includes a selection unit 71 and classification units 72-1 to 72 -N (N is a positive integer).

  The selection unit 71 selects any one of the classification units 72-1 to 72-N for each element of the type vector.

  The classification units 72-1 to 72-N classify the performers using different methods. For example, the classification unit 72-1 classifies the target performer by calculating the type vector of the target performer by the K-nearest Neighbor method represented by Expression (2). For example, the classifying unit 72-N is configured as a naive Bayes classifier, and classifies the target performer by calculating the type vector of the target performer according to Equation (3).

  That is, in this case, the value obtained by any one of the classification units 72-1 to 72-N according to the selection by the selection unit 71 is set as one of the elements of the type vector of the target performer. Thus, the type vector of the target performer is obtained, and the target performer is classified.

  Which one of the classification units 72-1 to 72-N is suitable for obtaining each element of the type vector of the target performer is learned in advance using learning data, and the selection unit 71 Depending on the learning result, one of the classifying units 72-1 to 72-N is selected for each element of the type vector of the target performer.

  Thus, for each element of the target performer's type vector, the value of that element is determined in a manner suitable for that element.

  Next, correction of the target performer classification will be described.

  FIG. 7 is a block diagram illustrating still another example of the configuration of the information processing apparatus 11 when correcting the classification of the target performers. In FIG. 7, the same parts as those shown in FIG. 1 are denoted by the same reference numerals, and the description thereof is omitted.

  When correcting the classification of the target performer, the information processing apparatus 11 includes a content information database 31, a performer feature vector generation unit 32, a classification unit 33, a performer type database 34, and a classification result correction unit 91.

  The classification result correction unit 91 corrects the classification of the target attribute value according to the classification bias of the classified attribute value that satisfies the predetermined condition. That is, the classification result correction unit 91 corrects the result of classification of the target performer by the classification unit 33.

  For example, the classification result correction unit 91 includes a broadcast time zone, an appearance ratio of each type for each broadcast station, a genre, an office to which the performer belongs, and a performer for content that is a program broadcast from the broadcast station. The result of the classification of the target performer by the classification unit 33 is corrected using the restriction condition 92 such as the age of the performer or the sex of the performer, and the classification is made more accurate.

  The classification result correcting unit 91 supplies the corrected performer classification result to the performer type database 34 or outputs it to the outside. The performer type database 34 holds the result of the classification of the modified performers.

  The process of creating the performer feature vector and the process of classification by the information processing apparatus 11 of FIG. 7 are the same as the processes described with reference to the flowcharts of FIGS. To do.

  FIG. 8 is a flowchart for explaining classification correction processing by the classification result correction unit 91. In step S <b> 51, the classification result correction unit 91 reads out the type vectors of all performers from the performer type database 34.

  In step S52, the classification result correction unit 91 determines whether the correction of the classification result has been completed for each time zone of the broadcasting station. Here, the broadcast station and the time zone are examples of the constraint condition 92.

  If it is determined in step S52 that the correction of the classification result has not been completed for each time zone of the broadcasting station, the procedure proceeds to step S53, and the classification result correcting unit 91 determines the broadcasting station and the time zone. Among them, one time zone of one broadcasting station that has not yet been selected for correcting the classification result is selected. In this case, for example, one broadcast station is selected from broadcast stations that broadcast programs that are contents, and one time slot is selected from a plurality of time slots obtained by dividing one day into a predetermined number. .

  In step S54, the classification result correcting unit 91 appears in a program broadcast from the selected broadcasting station in the selected time zone from among the type vectors of all performers read in step S51. Get the performer's type vector.

  In step S55, the classification result correction unit 91 determines whether or not the constraint condition 92 is satisfied.

  The classification result correction unit 91 may extract the constraint condition 92 before determining whether or not the constraint condition 92 is satisfied. That is, the classification result correction unit 91 obtains the appearance ratio of the performer type in each cluster as the constraint condition 92 in the cluster.

More specifically, for example, the classification result correction unit 91 uses each broadcast station and each time zone (0:00 to 3:00, 3:00 to 6) based on the data of the program guide for half a year or one year. : 00,6: 00 to 9: 00,9: 00 to 12: 00,12: 00 to 15: 00,15: 00 to 18: 00,18: 00 to 21:00, or 21:00 to 0: For each time zone such as 00), a program broadcast from each broadcasting station in each time zone is specified. Then, the classification result correction unit 91 stores in the performer type database 34 a restriction condition 92 indicating what type of performers are likely to appear in the programs broadcast from the respective broadcast stations in the respective time zones. From the type vector of the performer who has been classified as the stored learning data, it is obtained by equation (4).

  The constraint condition 92 obtained by the equation (4) indicates the bias of the type of the classified performer who appears in the program broadcast from the predetermined broadcasting station in the predetermined time zone.

Alternatively, the classification result correcting unit 91 specifies a program for each genre such as variety, news, or drama based on the data of the program guide for half a year or one year instead of the broadcasting station and the time zone. Then, the classification result correction unit 91 classifies the restriction condition 92 indicating what type of performers are likely to appear in the programs of each genre as learning data stored in the performer type database 34. From the type vector of the performed performer, it is obtained by equation (5).

  The restriction condition 92 obtained by the equation (5) indicates the type bias of the classified performers who appear in programs of a predetermined genre.

  Then, the classification result correction unit 91 selects a cluster that is effective for the restriction (a cluster that is effective for correction of the classification) from the restriction conditions 92 of the clusters such as the time zone, the broadcasting station, and the genre. For example, as a restriction condition 92 for a performer appearing in a program broadcast from a predetermined broadcast station during a time period from 18:00 to 21:00, among the types of classified performers, appearing in a comedy program The comedy type is 70%, the moderator type that hosts the program is 30%, and it appears in programs broadcast from a given broadcasting station in the 15:00 to 18:00 period As the restriction condition 92 for performers, among the types of performers that have already been classified, 55% are comedy-types appearing on comedy programs, and 45% are moderator-types that host programs. In some cases, a restriction condition 92 for a performer appearing in a program broadcast from a predetermined broadcast station in a time zone with a bias from 18:00 to 21:00 is selected.

  Then, the classification result correcting unit 91 determines whether or not the type appearance probability based on the type vector of the target performer is applicable to the selected restriction condition 92.

  For example, the classification result correction unit 91 calculates the appearance probability for each type for the target performers in the same cluster as the selected restriction condition 92. For example, the classification result correcting unit 91 replaces the type vector of the performer that has been classified, and the constraint selected by the calculation represented by the formula (4) or the formula (5) based on the type vector of the target performer. For the target performers in the same cluster as the condition 92, the appearance probability for each type is calculated.

  Then, the classification result correction unit 91 compares the selected restriction condition 92 with the selected restriction condition 92 by comparing the appearance probability for each type of the target performers in the same cluster as the selected restriction condition 92. Is determined, that is, whether the appearance probability for each type of the target performers in the same cluster as the selected constraint 92 is the same as the selected constraint 92 is determined.

  In step S55, when it is determined that the type appearance probability based on the type vector of the target performer does not apply to the selected restriction condition 92, the procedure proceeds to step S56, and the classification result correction unit 91 performs the restriction condition 92. The type vector of each target performer is corrected so that

For example, if the type vector of the target cast by the formula (2) is calculated, the classification result correction unit 91, by changing the threshold value of f _x, as is true in the constraint 92, the target actor's respectively Correct the type vector. More specifically, when the appearance probability of a predetermined type for the target performer is larger than the appearance probability of that type for the classified performer indicated by the restriction condition 92, the classification result correction unit 91 sets a threshold value. Is increased so that the type vector of the target performer is corrected so as to lower the appearance probability of that type for the target performer. On the other hand, when the appearance probability of the predetermined type for the target performer is smaller than the appearance probability of that type for the classified performer indicated by the restriction condition 92, the classification result correction unit 91 decreases the threshold value. Thus, each type vector of the target performer is corrected so as to increase the appearance probability of that type for the target performer.

  For example, among the classified performers who appear in programs broadcast from a predetermined broadcast station in the time period from 18:00 to 21:00, 30% of the classified performers are of the comedy type. If there are 70% of categorized performers who are romance types appearing in, among the target performers appearing in programs broadcast from that broadcast station in the time zone from 18:00 to 21:00, When the target performer classified as the type of comedy is 45% and the target performer classified as the type of love is 55%, the classification result correction unit 91 determines that the target performer is a comedy Each of the target performers is changed by changing the threshold so that the ratio of the target performer is classified as a romantic type and the threshold is changed so that the ratio of the target performer is reduced. Correct the type vector.

  That is, in step S56, the classification result correction unit 91 corrects the classification of the target attribute value according to the bias of the classification of the classified attribute value that satisfies the predetermined condition.

  The classification result correcting unit 91 not only changes the threshold value for classifying the target performer into a type, but also changes the threshold value for classifying into the type for all performers. Each type vector may be corrected.

  For example, there are many types of performers who are moderators in the program that are broadcast from 6:00 to 9:00, and they are appearances of a type that is a romantic type that appears in a love drama. Since the number of performers is small, the classification result correcting unit 91 has a value of an element indicating a romantic system among elements of type vectors of performers appearing in a program broadcast in a time zone from 6:00 to 9:00. If it is higher, the performer's type vector is modified so that the performer is categorized as a romance to indicate that it is a romance.

  In addition, for programs broadcast from each broadcasting station in each time zone, the difference between the appearance probability of the classified performer's classification and the appearance probability of the target performer's classification is taken as the error, and each of all time zones The type vector is modified so that the sum of errors for each of all broadcast stations is minimized.

  In this way, it is possible to classify more accurately by changing the threshold for classification and correcting the classification of performers so as to match the classification tendency of the classified performers indicated by the learning data given manually. Will be able to.

  After step S56, the procedure returns to step S52.

  If it is determined in step S55 that the constraint condition 92 is satisfied, the type vector does not need to be corrected, so step S56 is skipped and the procedure returns to step S52.

  If it is determined in step S52 that the correction of the classification result has been completed for each time zone of the broadcasting station, the classification correction process ends.

  Next, an example of more specific classification correction processing by the classification result correction unit 91 will be described with reference to the flowchart of FIG. In step S71, the classification result correction unit 91 determines whether or not the correction of the type vector is completed for each time zone of the broadcasting station.

  If it is determined in step S71 that the correction of the type vector has not been completed for each time zone of the broadcasting station, the procedure proceeds to step S72, and the classification result correcting unit 91 determines whether the broadcasting station and the time zone are One time zone of one broadcasting station is selected from among the combinations that have not yet been selected for correcting the classification result.

  In step S <b> 73, the classification result correcting unit 91 obtains from the performer type database 34 the type vectors of the classified performers who have appeared in the program broadcast from the selected broadcasting station in the selected time zone.

  In step S74, the classification result correcting unit 91 calculates the appearance ratio of the type in the selected time zone and the selected broadcasting station from the type vector of the classified performer.

  In step S75, the classification result correcting unit 91 determines whether or not the type appearance ratio has a bias equal to or greater than a predetermined threshold. For example, the classification result correction unit 91 subtracts the appearance ratio of the romantic type from the appearance ratio of the comedy type, compares the absolute value of the subtraction result with a predetermined threshold, By determining whether or not the absolute value of the result exceeds the threshold value, it is determined whether or not the appearance ratio of the type that is a comedy type and the type that is a romance type is more than a threshold value.

  If it is determined in step S75 that the type appearance ratio has a bias greater than or equal to a predetermined threshold, the procedure proceeds to step S76, and the classification result correction unit 91 selects the broadcast station selected in the selected time zone. From the performer type database 34, the type vector of the target performer appearing in the program broadcast from is acquired.

  In step S77, the classification result correcting unit 91 calculates the appearance ratio of the type in the selected time zone and the selected broadcasting station from the type vector of the target performer.

  In step S78, the classification result correcting unit 91 sets the appearance ratio of the type of the already-performed performer calculated in step S74 and the appearance ratio of the type of the target performer calculated in step S77 to a predetermined threshold value or more. It is determined whether there is a difference.

  For example, the classification result correction unit 91 determines whether or not there is a difference equal to or greater than a predetermined threshold value between the appearance ratio of the classified performer and the appearance ratio of the target performer of the same type.

  For example, more specifically, the classification result correcting unit 91 subtracts the appearance ratio of the type that is the comedy type of the target performer from the appearance ratio of the type that is the comedy type of the classified performer, and obtains the result of the subtraction. By comparing the absolute value with a predetermined threshold and determining whether or not the absolute value of the result of the subtraction exceeds the threshold, the appearance ratio of the type that is the comedy type of the classified performer and the target performer It is determined whether there is a difference equal to or greater than a threshold in the appearance ratio of a type that is a comedy type.

  If it is determined in step S78 that there is a difference equal to or greater than a predetermined threshold value between the appearance ratio of the type of the performer that has been classified and the appearance ratio of the type of the target performer, the procedure proceeds to step S79, and the classification result correction unit 91 corrects each type vector of the target performer so that the appearance ratio of the type of the target performer matches the appearance ratio of the type of the classified performer according to the appearance ratio of the type of the classified performer. .

  After step S79, the procedure returns to step S71 and repeats the above-described processing. Further, when it is determined in step S78 that there is no difference equal to or greater than a predetermined threshold value between the appearance ratio of the type of the classified performer and the appearance ratio of the type of the target performer, and in step S75, the appearance ratio of the type If it is determined that there is no bias greater than or equal to a predetermined threshold value, it is not necessary to correct each type vector of the target performer, so the procedure returns to step S71 and repeats the above-described processing.

  If it is determined in step S71 that the type vector correction has been completed for each time zone of the broadcast station, the classification correction process ends.

  In this way, the classification of the target performers can be corrected so that the classification becomes more accurate.

  Note that the restriction condition 92 may be not limited to the broadcasting station, time zone, or genre, but may be the affiliated office, age, gender, or the like.

  As described above, the classification result correction unit 91 corrects the classification of the target attribute value according to the bias of the classification of the attribute values already classified for the attributes that satisfy the predetermined condition among all the contents. .

  In addition, the classification result correction unit 91 responds to the bias in the classification of the attribute values that have been assigned to the content that satisfies the conditions that limit the attributes of the content in which the predetermined performer indicated by the target attribute value appears. , Correct the classification of the target attribute value.

  The classification result correction unit 91 is a classified attribute value indicating a performer whose content of the target attribute value indicating a predetermined performer who appears in content satisfying a predetermined condition satisfies the condition that satisfies the condition. The classification of the target attribute value is corrected so as to be closer to the bias of the classification.

  Furthermore, when the classification bias of the classified attribute value indicating the performer appearing in the content satisfying the predetermined condition is larger than a predetermined threshold, the classified attribute having a bias larger than the threshold is displayed. The classification of the target attribute value is modified according to the value classification.

  Furthermore, the classification result correction unit 91 has a classification bias of classified attribute values indicating performers appearing in content satisfying a predetermined condition larger than a predetermined first threshold and satisfies the condition The difference between the classification bias of the classified attribute value indicating the performer who appears in the content and the bias of the classification of the target attribute value indicating the performer appearing in the content satisfying the condition satisfies the predetermined second threshold value. If it is larger, the classification of the target attribute value is corrected according to the classification of the classified attribute value whose bias is larger than the first threshold.

  The classification result correction unit 91 can also correct the classification of the target attribute value according to the classification bias of the classified attribute value that satisfies the conditions for limiting other attributes of the performer indicated by the target attribute value.

  Further, the classification correction process is not limited to one time, and may be repeatedly executed. The amount of classification correction processing is small compared to the amount of performer feature vector creation or classification processing (because the calculation is light), so the classification correction processing is executed repeatedly. However, the load will not increase significantly.

  Furthermore, a classification unit may be provided for each element of the type vector so that the element of the target performer may be corrected after obtaining the element by a different method for each element of the type vector.

  FIG. 10 is a block diagram illustrating still another example of the configuration of the information processing apparatus 11 when a classification unit is provided for each element of the type vector and the classification of the target performer is corrected. 10, the same reference numerals are given to the same parts as those shown in FIG. 1, the case shown in FIG. 6, or the case shown in FIG. 7, and the description thereof will be omitted as appropriate.

  In this case, the information processing apparatus 11 includes a content information database 31, a performer feature vector generation unit 32, a performer type database 34, a selection classification unit 51, and a classification result correction unit 91.

  In the information processing apparatus 11 of FIG. 10, for each element of the type vector of the target performer, the value of the element is obtained in a method suitable for the element, and further, the classification by the type vector is more accurate. The target performer classification is modified.

  The process of creating a performer feature value vector, the process of classification, and the process of correcting the classification by the information processing apparatus 11 of FIG. 10 have been described with reference to the flowcharts of FIG. 2, FIG. 5, and FIG. Since it is the same as each processing, its description is omitted.

  As described above, the type can be automatically given to the performer by using the data given by a small number of hands as learning data.

  By doing in this way, the performer who has begun to appear in the program can be classified into the correct type without viewing the program in which the program appears.

  In addition, since the classification result is corrected using the constraint conditions specific to the content, the classification can be performed more accurately.

  For example, it is possible to classify performers based on the learning data generated by individual viewers with the performers of programs that viewers often watch as classified performers. Performers are classified according to preferences, and by specifying a viewer's favorite program based on this classification, the accuracy of recommending a program to the viewer is improved.

  The information processing apparatus 11 can be realized as a television receiver, a tuner, a video recorder player with a built-in HDD (Hard Disk Drive), a set top box, a server, or a terminal device as a client.

  Thus, when the feature amount of the performer is obtained and the classified performer is classified in advance, the performer can be classified from the feature amount of the performer and the classification of the classified performer. . In addition, a feature quantity indicating a feature of the attribute value of the attribute assigned to the content is created from the information about the content, and the target attribute value that is the attribute value to be classified is set to the same attribute as the attribute of the target attribute value. The attribute value is classified according to the classification of the classified attribute value of the feature quantity similar to the feature quantity of the target attribute value among the classified attribute values that are the attribute values that are classified in advance, and a predetermined condition If the classification of the target attribute value is modified according to the classification bias of the classified attribute value assigned to the content that satisfies the criteria, the attribute attribute assigned to the content can be more easily and accurately Values can be classified.

  The series of processes described above can be executed by hardware or can be executed by software. When a series of processing is executed by software, a program constituting the software executes various functions by installing a computer incorporated in dedicated hardware or various programs. For example, it is installed from a program recording medium in a general-purpose personal computer or the like.

  FIG. 11 is a block diagram showing an example of the configuration of a personal computer that executes the above-described series of processing by a program. A CPU (Central Processing Unit) 201 executes various processes according to a program stored in a ROM (Read Only Memory) 202 or a storage unit 208. A RAM (Random Access Memory) 203 appropriately stores programs executed by the CPU 201 and data. The CPU 201, ROM 202, and RAM 203 are connected to each other via a bus 204.

  An input / output interface 205 is also connected to the CPU 201 via the bus 204. Connected to the input / output interface 205 are an input unit 206 made up of a keyboard, mouse, microphone, and the like, and an output unit 207 made up of a display, a speaker, and the like. The CPU 201 executes various processes in response to commands input from the input unit 206. Then, the CPU 201 outputs the processing result to the output unit 207.

  A storage unit 208 connected to the input / output interface 205 includes, for example, a hard disk, and stores programs executed by the CPU 201 and various data. The communication unit 209 communicates with an external device via a network such as the Internet or a local area network.

  Further, a program may be acquired via the communication unit 209 and stored in the storage unit 208.

  The drive 210 connected to the input / output interface 205 drives a removable medium 211 such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, and drives the programs and data recorded therein. Get etc. The acquired program and data are transferred to and stored in the storage unit 208 as necessary.

  As shown in FIG. 11, a program recording medium that stores a program that is installed in a computer and is ready to be executed by the computer includes a magnetic disk (including a flexible disk), an optical disk (CD-ROM (Compact Disc-Read Only). Memory, DVD (Digital Versatile Disc), a magneto-optical disk, a removable medium 211 that is a package medium composed of a semiconductor memory, or the like, a ROM 202 in which a program is temporarily or permanently stored, or a storage unit 208 It is comprised by the hard disk etc. which comprise. The program is stored in the program recording medium using a wired or wireless communication medium such as a local area network, the Internet, or digital satellite broadcasting via a communication unit 209 that is an interface such as a router or a modem as necessary. Done.

  In the present specification, the step of describing the program stored in the program recording medium is not limited to the processing performed in time series in the described order, but is not necessarily performed in time series. Or the process performed separately is also included.

  The embodiment of the present invention is not limited to the above-described embodiment, and various modifications can be made without departing from the gist of the present invention.

It is a block diagram which shows an example of a structure of the information processing apparatus of one embodiment of this invention. It is a flowchart explaining the process of preparation of a performer feature-value vector. It is a figure which shows the example of content information. It is a figure which shows the example of a performer feature-value vector. It is a flowchart explaining the process of performer classification. FIG. 12 is a block diagram illustrating another example of the configuration of the information processing apparatus. FIG. 22 is a block diagram illustrating still another example of the configuration of the information processing device. It is a flowchart explaining the process of correction of classification. 10 is a flowchart for explaining an example of more specific classification correction processing. FIG. 22 is a block diagram illustrating still another example of the configuration of the information processing device. And FIG. 11 is a block diagram illustrating an example of a configuration of a personal computer.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 11 Information processing apparatus, 31 Content information database, 32 Performer feature-value vector preparation part, 33 Classification part, 34 Performer type database, 51 Selection classification part, 71 Selection part, 72-1 thru | or 72-N Classification part, 91 classification Result correction unit, 201 CPU, 202 ROM, 203 RAM, 208 storage unit, 211 removable media

Claims (13)

A feature amount creating means for creating a feature amount indicating a feature about an attribute value of an attribute given to the content from information about the content;
The target attribute value that is the attribute value to be classified is the attribute value of the same attribute as the attribute of the target attribute value, and the target attribute of the classified attribute values that are pre-classified attribute values First classification means for classifying according to the classification of the attribute value that has been classified similar to the feature value of the value;
An information processing apparatus comprising: correction means for correcting the classification of the target attribute value according to a classification bias of the classified attribute value assigned to the content that satisfies a predetermined condition. 2. The correction unit corrects the classification of the target attribute value according to a bias in the classification of the classified attribute value of the attribute assigned to the content that satisfies the condition among all the contents. The information processing apparatus described in 1. The correction means responds to a bias in the classification of the classified attribute value of the attribute given to the content that satisfies the condition satisfying the condition that limits the attribute of the content in which the predetermined performer indicated by the target attribute value appears The information processing apparatus according to claim 2, wherein the classification of the target attribute value is corrected. The correction means is the classified already indicating that the target attribute value classification bias indicating a predetermined performer appearing in the content satisfying the condition satisfies a performer appearing in the content satisfying the condition The information processing apparatus according to claim 3, wherein the classification of the target attribute value is corrected so as to be closer to the bias of the attribute value classification. When the bias of classification of the classified attribute value indicating a performer who appears in the content that satisfies the condition satisfies a bias that is greater than the first threshold, the correction means has a bias that is greater than the first threshold. The information processing apparatus according to claim 3, wherein the classification of the target attribute value is modified according to the classification of the classified attribute value. The correction means has a classification bias of the classified attribute value indicating a performer appearing in the content that satisfies the condition that is greater than a predetermined first threshold, and appears in the content that satisfies the condition A difference between a bias in the classification of the classified attribute value indicating the performer performing and a bias in the classification of the target attribute value indicating the performer performing in the content that satisfies the condition is determined in advance. The information processing apparatus according to claim 3, wherein when the threshold value is larger than the threshold value, the classification of the target attribute value is corrected according to the classification of the classified attribute value biased to be larger than the first threshold value. The first classification unit is configured to determine the classified attribute value of the feature amount that is similar to the feature amount of the target attribute value based on a similarity weighted for each element of the feature amount including a plurality of elements. The information processing apparatus according to claim 1, wherein the target attribute value is classified according to the classification. The first classification unit is configured to classify the feature amount similar to the feature amount of the target attribute value indicating a predetermined performer among a plurality of types into the first type of the classified attribute value. The information processing apparatus according to claim 1, wherein the target attribute value indicating the performer is classified into the first type. The classification attribute value of the feature quantity similar to the feature quantity of the target attribute value indicating the performer among the plurality of types by a scheme different from the classification scheme of the first classification means. Second classification means for classifying the target attribute value indicating the performer into the second type according to the classification into two types;
The information processing apparatus according to claim 8, further comprising: a selection unit that selects one of the first classification unit and the second classification unit for each type. The feature amount creating means creates the feature amount indicating a feature about the attribute value indicating a performer who appears in the content that is a program from information about the content that is a broadcast program,
The correction means is configured to set the target attribute value in accordance with a classification bias of the classified attribute value indicating a performer who appears in the content that is a program broadcast from a predetermined broadcast station in a predetermined time zone. The information processing apparatus according to claim 1, wherein the classification is corrected. Create a feature amount indicating a feature about the attribute value of the attribute given to the content from the information about the content,
The target attribute value that is the attribute value to be classified is the attribute value of the same attribute as the attribute of the target attribute value, and the target attribute of the classified attribute values that are pre-classified attribute values Classifying according to the classification of the attribute value classified attribute value similar to the feature value of the value,
An information processing method including a step of correcting a classification of the target attribute value according to a classification bias of the classified attribute value given to the content satisfying a predetermined condition. Create a feature amount indicating a feature about the attribute value of the attribute given to the content from the information about the content,
The target attribute value that is the attribute value to be classified is the attribute value of the same attribute as the attribute of the target attribute value, and the target attribute of the classified attribute values that are pre-classified attribute values Classifying according to the classification of the attribute value classified attribute value similar to the feature value of the value,
A program that causes a computer to execute a step of correcting the classification of the target attribute value in accordance with the classification bias of the classified attribute value given to the content that satisfies a predetermined condition. Create a feature amount indicating a feature about the attribute value of the attribute given to the content from the information about the content,
The target attribute value that is the attribute value to be classified is the attribute value of the same attribute as the attribute of the target attribute value, and the target attribute of the classified attribute values that are pre-classified attribute values Classifying according to the classification of the attribute value classified attribute value similar to the feature value of the value,
A program for causing a computer to execute information processing including a step of correcting the classification of the target attribute value according to the classification bias of the classified attribute value assigned to the content satisfying a predetermined condition is recorded. recoding media.

Images