JP2013003893A - Content relevance calculation device, content relevance calculation method and content relevance calculation program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To recommend a more appropriate content.SOLUTION: A content relevance calculation device 10 includes: a content designation part 11 for designating a content as a calculation object of relevance; a genre system storage part 17 for storing a genre system expressed as a hierarchical structure; a relevance score storage part 18 for storing relevance scores showing the strength of inter-genre relationship; a genre relevance calculation part 13 for specifying the closest ancestor which is common to the genres of the content designated by the content designation part 11 in the hierarchical structure stored in the genre system storage part 17, and for, when the relevance scores stored in the relevance score storage part 18 is defined in the hierarchy which is deeper than the ancestor, calculating the inter-genre relevance on the basis of the relevance scores; and a content relevance calculation part 14 for calculating inter-content relevance designated by the content designation part 11 on the basis of the inter-genre relevance calculated by the genre relevance calculation part 13.

Description

  The present invention relates to a content relevance calculation device, a content relevance calculation method, and a content relevance calculation program for calculating a relevance between contents.

  There is known a recommendation technique for increasing the number of accesses and sales of content by presenting other content to a user who has accessed the content. As one of the recommendation techniques, there is a method of calculating the relevance level of content by matching metadata attached to the content. Such a technique is called content-based filtering.

  In content-based filtering, we use information such as keywords and features extracted from content, tags attached to content, and attributes of content, and we believe that the degree of relevance of content that has more information in common is high. Make a recommendation. As research on content-based filtering, there is a method of measuring the frequency of appearance of attributes assigned to content highly evaluated by the user and recommending content to which attributes with high appearance frequency are assigned (for example, Patent Document 1). ).

JP 2009-205418 A

  Here, consider a case where video is recommended by a content-based filtering recommendation method. For example, when a user views a video A having a genre of “anime for children”, the user is likely to like another video for “anime for children”. Therefore, in the content-based filtering recommendation method, the degree of relevance between the video A and another “children's animation” video is calculated to be high. Also in the recommendation method disclosed in Patent Document 1, when the user is watching a lot of “anime for children”, the recommendation score of the video belonging to “anime for children” is calculated high.

  Users who watch “anime for children” are often children, and it is highly likely that they prefer videos of genres such as “puppet show” and “special effects”. It is considered effective to recommend images such as “SFX” and “Special effects”. On the other hand, video of genres such as “politics” and “horror” is likely not to be preferred, so it is better not to recommend it.

  However, in the conventional content-based filtering recommendation method disclosed in Patent Document 1 or the like, the recommendation score is not increased or decreased for a video of a genre that is not viewed by the user. For this reason, there are problems that videos of the “puppet show” and “special effects” genres that should be recommended are not recommended, and videos of “politics” and “horror” that should not be recommended.

  An object of the present invention is to provide a content relevance calculation device, a content relevance calculation method, and a content relevance calculation program capable of recommending more appropriate content.

  In order to solve the above-described problem, a content relevance calculation device according to an embodiment of the present invention is a device that calculates a relevance between contents, and a content specification unit that specifies content for which the relevance is to be calculated; A genre system accumulating unit for accumulating a genre system expressed as a hierarchical structure, a relationship score accumulating unit for accumulating a relationship score representing the strength of the relationship between genres, and the genre system accumulating unit. The closest ancestor common to the genre of the content specified by the content specifying unit in the hierarchical structure is defined, and the relationship score stored in the relationship score storage unit is defined in a hierarchy deeper than the ancestor A genre relevance calculating unit that calculates a relevance between genres based on the relevance score, and the genre relevance Based on the relevance between calculated by output unit genre; and a content relevance calculator that calculates a degree of association between the designated content by the content specifying unit.

  If the closest ancestor common to genre a and genre b is genre c, the genre relevance calculating unit genre a 'is an ancestor of genre a and deeper than genre c, and genre b' is an ancestor of genre b and deeper than genre c. When the relationship score is defined between the genre a ′, the higher the absolute value of the relationship score, the higher the degree of relevance, and the higher the genre a ′ and b ′, the higher the degree of relevance. May be.

The genre relevance calculating unit sets a relationship score defined between genres a ′ and b ′ as C (a ′, b ′), and how detailed are the genres a ′ and b ′. Let T (a ′, b ′) be a function that calculates a score based on, and the relationship score defined at a deeper level than the closest common ancestor in genre a and genre b is G ′ = {(a ′ _{i ,} B ′ _i ) | i = 1, 2,...}, The relation R _G (a, b) between the genre a and the genre b is expressed as R _G You may calculate by '(G').

  If the closest ancestor common to genre a and genre b is genre c, the genre relevance calculating unit genre a 'is an ancestor of genre a and deeper than genre c, and genre b' is an ancestor of genre b and deeper than genre c. If the relationship score is not defined between the two, the higher the degree of relevance may be calculated as the genre hierarchy corresponding to the ancestor is deeper.

When the function for calculating a score based on how detailed the genre c is is a genre c, the genre relevance calculation unit calculates the relevance R _G (a, b) between the genre a and the genre b. ) May be calculated by R _G ″ (c) in the following equation.

  In order to solve the above-mentioned problem, a content relevance calculation method according to an embodiment of the present invention is a method for calculating a relevance between contents, and a content designating step for designating content for which a relevance is to be calculated; A genre system accumulating step for accumulating a genre system expressed as a hierarchical structure, a relationship score accumulating step for accumulating a relationship score representing the strength of the relationship between genres, and the genre system accumulating step. In the hierarchical structure, the closest ancestor common to the content genre specified in the content specifying step is specified, and the relationship score stored in the relationship score storing step is defined in a hierarchy deeper than the ancestor , Genre relevance calculation that calculates the relevance between genres based on the relationship score And a content relevance level calculating step for calculating a relevance level between the contents specified in the content specifying step based on the relevance level between the genres calculated in the genre relevance level calculating step. .

  In order to solve the above-described problem, a content relevance calculation program according to an embodiment of the present invention causes a computer to function as each processing unit included in the content relevance calculation device.

  According to the present invention, since the relationship between genres is considered, a content relevance calculation device, a content relevance calculation method, and a content relevance calculation program capable of recommending more appropriate content are provided. can do.

It is a block diagram of the content relevance calculation apparatus in embodiment of this invention. It is explanatory drawing of the genre system in embodiment of this invention. It is a figure which shows an example of the content information in embodiment of this invention. It is a flowchart of the content relevance calculation apparatus in embodiment of this invention. It is a figure which shows an example of the relationship score in embodiment of this invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

  FIG. 1 is a configuration diagram of a content relevance calculation device 10 according to an embodiment of the present invention. The content relevance calculation device 10 is a device that calculates the relevance between contents, and as shown in FIG. 1, a content designation unit 11, a genre pair extraction unit 12, a genre relevance calculation unit 13, A content relevance calculation unit 14, a content relevance output unit 15, a content information storage unit 16, a genre system storage unit 17, and a relationship score storage unit 18 are provided. The content information storage unit 16 is a storage device or the like that stores content information. The genre system storage unit 17 is a storage device that stores a genre system expressed as a tree structure. The relationship score storage unit 18 is a storage device or the like that stores a relationship score representing the strength of the relationship between genres. The content designation unit 11 is an input device or the like that designates two contents that are targets of calculation of relevance. The genre pair extraction unit 12 extracts genre pairs for the two contents specified by the content specification unit 11 based on the content information stored in the content information storage unit 16. The genre association degree calculation unit 13 calculates the degree of association between genres for the genre pair extracted by the genre pair extraction unit 12. For example, when the closest ancestor common to a genre pair in a tree structure is specified and a relationship score is defined in a hierarchy deeper than the ancestor, the degree of association between genres is calculated based on the relationship score. The content relevance calculation unit 14 calculates the relevance between the two contents specified by the content specification unit 11 based on the relevance between genres calculated by the genre relevance calculation unit 13. The content relevance output unit 15 outputs the relevance between the two contents calculated by the content relevance calculation unit 14.

  FIG. 2 is an explanatory diagram of the genre system stored in the genre system storage unit 17. As shown in this figure, it is assumed that the genre system is defined as a tree structure. At this time, the genre system in different services may be expressed as one tree structure, and it is assumed that the root genre is arranged in the uppermost layer and the genre of each service is arranged in the lower layer. Here, the genre for which the relationship score is defined is indicated by a double arrow. The relationship score can be specified independently by the user. When a relationship is defined between genres corresponding to the ancestors, descendants, and siblings of each content genre, or with itself, the degree of association is calculated based on the relationship (described later).

FIG. 3 is a diagram illustrating an example of content information stored in the content information storage unit 16. As shown in the figure, a plurality of genres may be assigned to one content. The genre assigned to the content A is G _a = {a _i | i = 1, 2,...}, And the genre assigned to the content B is G _b = {b _j | j = 1, 2,. ..., The degree of association R (A, B) between content A and content B is calculated by the following equation.

Here, R _G (a _i , b _j ) is the degree of association between genre a _i and genre b _j . When one or more genres are not assigned to the content A or the content B, R (A, B) = 0 is set.

  FIG. 4 is a flowchart showing the operation of the content relevance calculation apparatus 10. Hereinafter, an operation when presenting another content to a user who accesses a certain content will be described.

  First, two contents for which the degree of association is to be calculated are designated (step S1). For example, one of the two contents is content accessed by the user, and the other is arbitrary content stored in the content information storage unit 16.

  Next, Steps S2 to S6 are repeated for an arbitrary genre pair assigned to each content. For example, genre c is the closest ancestor common to genre a and genre b. At this time, it is determined whether or not one or more relationship scores are defined between the genre a ′ that is an ancestor of the genre a and deeper than the genre c and the genre b ′ that is an ancestor of genre b and deeper than the genre c. (Step S2). However, the genre a ′ may be the genre a, and the genre b ′ may be the genre b. If the relationship score is defined, the degree of association between the genre a and the genre b is calculated based on the relationship score according to Equation 4 described later (step S3). On the other hand, when the relationship score is not defined, the degree of association between genre a and genre b is calculated based on the depth of the common ancestor by Equation 7 described later (step S4).

  Next, among the calculated relevance levels between genres, X is set to be the largest (steps S5 → S6). That is, when the calculated degree of association between genres is greater than X, or when X is undefined, the calculated degree of association between genres is X.

  Next, based on this X, the degree of association between the two contents is calculated by Equation 3, and the calculated degree of association is output (steps S7 → S8). If such processing is repeated, it is possible to calculate the degree of relevance of the content accessed by the user for all the content stored in the content information storage unit 16.

Hereinafter, the case where the relationship score is defined in a hierarchy deeper than the nearest ancestor common to the genre a and the genre b will be described in detail. First, it is assumed that the relationship is defined between the i-th genres in G ′ = {(a ′ _i, b ′ _i ) | i = 1, 2,. In this case, the genre relevance calculation unit 13 calculates the relevance R _G (a, b) between the genre a and the genre b by R _G ′ (G ′) in the following equation.

  C (a ′, b ′) is a relationship score defined between the genres a ′ and b ′. The relation score takes an arbitrary real value. A positive value is defined when the content of the genre b 'is recommended for the content of the genre a', a negative value when the recommendation is not desired, and a value is defined as 0 when neither is recommended. The higher the absolute value of the relationship score, the higher the degree of relevance is calculated. For example, in order to reduce the degree of association between “for children” and “horror”, C (for children, horror) = − 1 is defined as shown in FIG.

T (a ′, b ′) is a function for calculating a score based on how detailed the genres a ′ and b ′ are, and the larger the genre of a ′ and b ′, the larger the value. Become. For example, it is assumed that the genre a is “anime for children” and the genre b is “horror”. In this case, the closest ancestor common to the genre a and the genre b is “VOD”, and a relationship score is defined between the deeper levels “for children” and “horror”. At this time, the genre a ′ is “for children” and the genre b ′ is “horror”. T (a ′, b ′) can be calculated by the following equation, for example.

However, N (a ′) is a function that returns the number of contents belonging to the genre a ′ among all contents, and N _root is the number of all contents. α is a parameter, and an arbitrary value can be set. Assuming that N _root = 12000, α = 100, N (for children) = 1000, and N (horror) = 90, T (for children, horror) = 0.4. Therefore, the degree of association R _G between “anime for children” and “horror” is R _G ′ (for children, horror) = − 1 × 0.4 = −0.4.

Another method for calculating T (a ′, b ′) will be described. For example, T (a ′, b ′) may be calculated according to the number of genres a ′ and b ′ belong to from the end. At this time, T (a ′, b ′) can be calculated by the following equation.

  However, D (a ′) is a function that returns the number of levels that the genre a ′ belongs to from the end in the tree structure of the genre system, and D (for children) = 2. β is a parameter, and an arbitrary value can be set. If β = 3, D (for children) = 2, and D (horror) = 1, then T (for children, horror) = 1.

Next, the case where the relationship score is not defined in a hierarchy deeper than the nearest ancestor common to the genre a and the genre b will be described in detail. The nearest ancestor common to genre a and genre b is defined as genre c. In this case, the genre relevance calculation unit 13 calculates the relevance R _G (a, b) between the genre a and the genre b by R _G ″ (c) of the following equation.

The value of T (c, c) can be calculated by Equation 5 or Equation 6. For example, it is assumed that the genre a is “foreign film-SF” and the genre b is “Japanese film-action”. In this case, the closest ancestor common to genre a and genre b is “movie”, and no relationship score is defined in a deeper hierarchy. Therefore, _assuming that N _root = 12000, α = 100, and N (movie) = 1000, T (c, c) can be calculated as R _G ″ (movie) = 0.12 using Equation 5.

  As described above, according to the content relevance calculation device 10, since the relationship between genres is considered, more appropriate content can be recommended. That is, since the degree of association between contents can be calculated based on the relationship between the genres of each content even for two contents of different genres, it is more strongly related in the recommendation based on the recommendation technique of content-based filtering. More content can be recommended.

  Further, according to the content relevance calculation device 10, when the relevance score is defined, the relevance between genres is calculated by Equation 4. In other words, the higher the absolute value of the relationship score, the higher the degree of association is calculated, and the higher the degree of genre a ′ and b ′ is, the higher the degree of association is calculated. Can be calculated.

  Further, according to the content relevance calculation device 10, when the relevance score is not defined, the relevance between genres is calculated by Equation 7. That is, since the higher the degree of association is calculated as the level of the genre corresponding to the ancestor is deeper, the degree of association between genres can be appropriately calculated.

  In addition, this invention is not limited to the said embodiment, A change and application are possible within a claim. For example, the genre system storage unit 17 stores the genre system expressed as a tree structure. However, the genre system only needs to be expressed as a hierarchical structure, and is not limited to the tree structure. In addition, although the user designates the relationship score, a person other than the user (for example, a content provider) can also designate the relationship score.

  In addition, the present invention can be realized not only as the content relevance calculation device 10, but also as an image processing method using a characteristic processing unit included in the content relevance calculation device 10 as a step, It can also be realized as a content relevance calculation program that causes a computer to function as a unit. It goes without saying that such a program can be distributed via a recording medium such as a CD-ROM or a transmission medium such as the Internet.

DESCRIPTION OF SYMBOLS 10 ... Content relevance calculation apparatus 11 ... Content designation | designated part 13 ... Genre relevance degree calculation part 14 ... Content relevance degree calculation part 17 ... Genre system storage part 18 ... Relevance score storage part

Claims (7)

A content relevance calculation device that calculates relevance between contents,
A content designating unit for designating content for which relevance is to be calculated;
A genre system storage unit that stores genre systems expressed as a hierarchical structure;
A relationship score accumulating unit for accumulating a relationship score representing the strength of the relationship between genres;
In the hierarchical structure stored in the genre system storage unit, the closest ancestor common to the genre of the content specified by the content specification unit is specified, and stored in the relationship score storage unit in a hierarchy deeper than the ancestor If a relationship score is defined, a genre relevance calculating unit that calculates a relevance between genres based on the relationship score;
A content relevance calculating unit that calculates a relevance between contents specified by the content specifying unit based on a relevance between genres calculated by the genre relevance calculating unit;
A content relevance calculation device comprising:   If the closest ancestor common to genre a and genre b is genre c, the genre relevance calculating unit genre a 'is an ancestor of genre a and deeper than genre c, and genre b' is an ancestor of genre b and deeper than genre c. If the relationship score is defined between and the higher the absolute value of the relationship score, the higher the degree of relevance is calculated, and the higher the degree of genre a ′ and b ′ is, the higher the degree of relevance is calculated The content relevance calculating apparatus according to claim 1, wherein: The genre relevance calculating unit sets a relationship score defined between genres a ′ and b ′ as C (a ′, b ′), and how detailed are the genres a ′ and b ′. Let T (a ′, b ′) be a function that calculates a score based on, and the relationship score defined at a deeper level than the closest common ancestor in genre a and genre b is G ′ = {(a ′ _{i ,} B ′ _i ) | i = 1, 2,...}, The relation R _G (a, b) between the genre a and the genre b is expressed as R _G The content relevance calculating apparatus according to claim 2, wherein the content relevance calculating apparatus calculates by “(G ′)”.

  If the closest ancestor common to genre a and genre b is genre c, the genre relevance calculating unit genre a 'is an ancestor of genre a and deeper than genre c, and genre b' is an ancestor of genre b and deeper than genre c. 2. The content relevance calculation apparatus according to claim 1, wherein when the relationship score is not defined between the two, the relevance degree is higher as the genre hierarchy corresponding to the ancestor is deeper. When the function for calculating a score based on how detailed the genre c is is a genre c, the genre relevance calculation unit calculates the relevance R _G (a, b) between the genre a and the genre b. ) Is calculated by R _G ″ (c) in the following equation.

A content relevance calculation method for calculating relevance between contents,
A content designating step for designating content for which relevance is to be calculated;
A genre system accumulating step for accumulating a genre system expressed as a hierarchical structure;
A relationship score accumulation step for accumulating a relationship score representing the strength of the relationship between genres;
In the hierarchical structure accumulated in the genre system accumulation step, the closest ancestor common to the genre of the content designated in the content designation step is specified, and the relation accumulated in the relationship score accumulation step in a hierarchy deeper than the ancestor If a sex score is defined, a genre relevance calculating step for calculating a relevance between genres based on the relevance score;
A content relevance calculating step for calculating a relevance between the contents specified in the content specifying step based on the relevance between the genres calculated in the genre relevance calculating step;
A content relevance calculation method characterized by comprising:   A content relevance calculation program for causing a computer to function as each processing unit according to any one of claims 1 to 5.

Images