JP5234836B2 - Content management apparatus, information relevance calculation method, and information relevance calculation program - Google Patents

Description

  The present invention relates to an information relevance calculation technique for calculating the relevance between annotations given to content and the relevance between annotations and content in a system for managing content such as documents, images, music, and videos. Is.

  In services that handle content such as digitized documents, images, music, and videos, words called tags and short sentences are added to the content as annotations and used for classification and search.

  The method of classifying content using annotation is different from the conventional general method of classifying one content into one category. By using tags, etc., multiple attribute information can be assigned to one content. It can be given and classified. In this way, multiple classification information can be added to content with multiple attributes and characteristics, and the user can search and narrow down content by specifying multiple attributes and characteristics. Become.

  Services that use tags for content classification and search include Hatena (registered trademark) bookmarks and Nico Nico Douga (registered trademark) in Japan, and YouTube (registered trademark), Flickr (registered trademark), and delicious in other countries. To do. In these services, the user can freely refer to the tag added to each content. At this time, the tags are generally arranged in order from the earliest date and time added to the content and presented to the user.

  Furthermore, in delicious, the tag group attached to the web page which is the content is classified and displayed according to which user attached, and since many users attached all the tags attached to the web page. A technique of arranging and displaying in order is used. That is, tags added by more users are placed higher.

  As an existing search technique for calculating the degree of association between a tag and a content, there is a technique for ranking using a feature amount of an image (Non Patent Literature 1).

D. Liu, X. S. Hua, L. J. Yang, M. Wang and H. J. Zhang, “Tag Ranking”, In Proceedings of A-CM International World Wide Web Conference, Pages: 351-360, 2009.

  However, in the prior art, the annotation does not have a value indicating the degree of association with the content, and is not ordered according to the degree of the degree of association with the content. As a result, users cannot determine which annotations typically represent content, and system developers make rankings using the degree of association between annotations and content and recommend related content The system cannot be created.

  The above-described conventional technology such as delicious can be applied only to a system in which tags added to content are classified for each user. Furthermore, in ranking based on the number of tags, general tags that are commonly added by many users such as category names tend to be higher. For this reason, tags that can be estimated by narrowing down the characteristics of the content from the tag name are not necessarily ranked higher.

In Non-Patent Document 1, the degree of association between a tag and an image is calculated using the feature amount of the image, the similarity between the image at the tag t _j and the tag t _j, and Google (registered trademark) of the following formula (1) The degree of relevance is corrected using the distance between the tags defined by the distance. Therefore, it is a method that can be applied only to an image, and costs such as analysis time and calculation amount are required as compared with the case where only annotation information is used.

d (t _i , t _j ) = {max (log f (t _i ), log f (t _j )) − log f (t _i , t _j )} ÷ {log G−min (log f (t _i )) , Log f (t _j ))} (1)
Here, f (t _i ) and f (t _j ) are the number of images to which tag t _i and tag t _j are added, and f (t _i , t _j ) are f (t _i ) and f (t _j ). ) Are added together, and G is the total number of images.

  The present invention is for solving the above-mentioned problems, and the degree of association between annotation and content is calculated by using only annotation information for annotation added for the purpose of content classification and the like. It is characterized by what it can do. The calculated relevance can be used for content ranking and recommendation.

  In order to solve the above-described problem, the present invention relates to a content management apparatus having an information storage unit storing annotations accompanying content with respect to a plurality of annotations added to a single content depending on the degree related to the content. Is a method for calculating the degree of annotation, which is obtained by acquiring an annotation group added to each content, calculating the degree of association between arbitrary annotations (hereinafter referred to as A-A degree of association), and the relation between the calculated annotations. The degree of association between an annotation and another annotation group is calculated using the degree, and the degree of association between the annotation and the content (hereinafter referred to as AC relation degree) is calculated and ranked.

  The calculation of the degree of association between annotations is based on the content rate of the content set with another annotation added to the content set with a certain annotation, the co-occurrence frequency in the same content of the annotation, or the appearance frequency of the annotation. Independence test values, etc. are available.

  In addition, in calculating the degree of association between an annotation and another annotation group added to the same content, the sum of the degrees of AA association with other annotation groups, or the AA association with other annotation groups A random surfer model in which the median degree or the ratio of AA relevance with other annotation groups is used as the link strength can be used.

  Both the invention described in Non-Patent Document 1 and the present invention match in that the relationship between the annotation and the content is obtained. However, there are significant differences in the following points. In the technique described in Non-Patent Document 1, in addition to analyzing the relationship between annotations, an image feature that is a feature of the content itself is obtained, and the relationship between the annotation and the content is calculated. On the other hand, the present invention can calculate the relationship between the annotation and the content based only on the relationship between the annotations.

  The reason why the present invention can determine the relationship between the annotation and the content only by the relationship between the annotations will be described with reference to FIG. FIG. 1 is an image diagram of content contents described by annotation.

  Annotations are added by capturing the content. Therefore, it can be said that each annotation expresses at least a part of the content. In the example of FIG. 1, “Kanagawa” to which six annotations (tags) such as “travel”, “Kanagawa prefecture”, “train”, “Enoden”, “Enoshima”, and “Kamakura” are added as shown in FIG. For the video content titled “Prefectural Tourist Attractions”, the part represented by each annotation is shown for the entire video content.

The present invention is based on the following knowledge.
[Knowledge 1]: The content that is duplicated by more annotations is the main content.

  Many annotations will try to express particularly important content in the whole content. In FIG. 1, an area where many circles indicating the contents represented by the annotation overlap each other corresponds to the “main contents” area.

Further, the present invention obtains the relationship between each annotation and content based on the following knowledge.
[Knowledge 2]: Annotations that express the main contents are highly related to the contents.

  In FIG. 1, the larger the area where more circles of annotation are overlapped, the higher the relevance with the content.

  Therefore, the main contents of the content can be estimated by estimating how much the annotations overlap.

  In the present invention, the degree of association between annotations indicating how much information is expressed with information of other annotations added to the same content is calculated. An asymmetric relationship between annotations can be defined by using a means for calculating a degree of association based on the content rate of a content set to which another annotation is added to a content set to which a certain annotation is added. Alternatively, by using a means for calculating the degree of association based on the co-occurrence frequency in the same content of the annotation, the degree of relevance that is not easily affected by the difference in the number of contents with the annotation added can be calculated. Alternatively, by using a means for calculating the degree of association based on the value of the independence test for the appearance frequency of the annotation, it is possible to calculate the degree of association with high accuracy using statistical information.

  In the present invention, the degree of association between the annotation and the content is calculated based on how many annotations common to the same content can be expressed among other annotations added to the same content. By using a means for calculating the degree of association based on the sum of the degrees of AA association with other annotation groups, the degree of association between the annotation and the content can be calculated with a small calculation cost. Alternatively, by using a means for calculating the relevance level based on the median AA relevance level with other annotation groups, it is possible to calculate the relevance level that is not easily affected by an outlier of the AA relevance level. Alternatively, by using a means for calculating the degree of association based on a random surfer model with the link strength as the ratio of AA relevance with other annotation groups, the relationship with all tags attached to the same content is considered. Highly accurate relevance can be calculated.

  Therefore, since the method of the present invention uses only annotation information and does not use information that depends on the type of content such as image feature amount, it can be applied to any content with an annotation added. It is different from the conventional technology in that the calculation cost is low compared with the method that uses the.

  Conventionally, annotation groups added to content are generally presented in the order in which annotations are added. For this reason, it is difficult to know which annotations added to the content express the features of the content more appropriately and in detail, and it was not possible to grasp the outline of the content at a glance.

  On the other hand, according to the present invention, it is possible to calculate the degree of association with content for each annotation added to the content and arrange the annotations in descending order of degree of association with the content. Therefore, it is possible to grasp the outline of the content only by looking at the annotation with high relevance.

  In addition, the search result list in the conventional search specifying annotations could only be sorted by an index that is not related to the degree of association between annotations and content, such as the number of times viewed or the date and time of posting. By using the present invention, it becomes possible to arrange and present the search result contents in descending order of the degree of association with the annotation specified as the search condition.

  Furthermore, in the present invention, by finding other contents including main annotations in a certain content, it is possible to obtain related contents with higher accuracy than the conventional method of finding contents having a large number of overlapping annotations. Is possible. In other words, in the conventional method, the annotations that are not able to express the characteristics of the content are duplicated in the annotations added to the content, and the case where the typical annotations are duplicated for the content is treated equally. On the other hand, by using the present invention, it is possible to acquire content in which more typical annotations are duplicated more.

  As described above, according to the present invention, by calculating the degree of association between content and annotations added to the content, information that is extremely useful for content search, content recommendation, etc., which cannot be obtained by the prior art. Will be able to provide.

It is an image figure of the contents contents described by annotation. It is a figure which shows the structural example of the content management apparatus which concerns on embodiment of this invention. It is a process flowchart of an information relevance calculation part. It is a figure which shows the example of the annotation information stored in the annotation storage apparatus. It is a figure which shows the example of the annotation-content relationship management table memorize | stored in A-C relevance degree memory | storage device. It is a figure which shows the example of the AA relevance degree about the annotation added to the content. It is a figure which shows the example of the content to which the annotation was added. It is a schematic diagram which shows the example of the calculation method of AA relevance degree. It is a schematic diagram which shows the example of the AC relevance calculation method from AA relevance. It is a figure which shows the structural example of the content management apparatus which rearranges a content set. It is a figure which shows the structural example of the content management apparatus which acquires a related content set.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 2 is a block diagram schematically showing the content management apparatus according to the embodiment of the present invention.

  The content management apparatus 10 includes an annotation storage apparatus 100 that holds content and annotations, an information relevance calculation unit 110, a relevance between the annotation calculated by the information relevance calculation unit 110 and the content (A-C relevance). A-C relevance degree storage device 120 is stored. The information relevance calculation unit 110 includes a content selection unit 111, an AA relevance calculation unit 112, and an AC relevance calculation unit 113.

  In this example, the annotation storage apparatus 100 stores a content information management table 101 and an annotation information management table 102. The A-C relevance degree storage device 120 stores an A-C relation management table 121 that holds information on the strength (relevance degree) of the relation between annotations and contents. In order to make the explanation easy to understand, in this example, the annotation storage device 100 and the AC relevance degree storage device 120 are shown as separate devices, but they may be the same storage device. Although the description is given here assuming that the content itself is also stored in the annotation storage device 100, the annotation storage device 100 manages only the content ID for specifying the content and the annotation group, and the content itself is the other. It may be stored in the device.

  The input / output device 20 is a display, a keyboard, or other peripheral devices used by a person who performs a relevance calculation operation, but may be a system that uses the relevance. Hereinafter, a person or system that uses the content management apparatus 10 is referred to as a “user”.

  The content management apparatus 10 includes a program developed in storage means such as a central processing unit (CPU), a read only memory (ROM), a random access memory (RAM), a hard disk drive (HDD), and a RAM. The program describing the operation of the components shown in FIG. 2 can be executed on a computer used as the content management apparatus 10, or can be executed as a service via a network or the like. The annotation storage apparatus 100 may be a program that acquires content and annotation through an API (Application Program Interface) or the like.

  The information relevance calculation unit 110 calculates an AC relevance level that represents the strength of the relationship between each annotation and the content as a numerical value for a plurality of annotations added to a certain content. Here, the content includes, for example, an electronic document, image, music, video, and the like, and the annotation includes a tag, a keyword, metadata, user information, and the like.

  That is, the information relevance level selection unit 110 determines the content whose relevance level is to be calculated in the content selection unit 111 in response to a request from the user such as the input / output device 20 or another system. The A relevance calculation unit 112 calculates the AA relevance, and the AC relevance calculation unit 113 calculates the AC relevance. Output. Alternatively, the calculated AC relevance degree is stored in the AC relevance degree storage device 120.

  FIG. 3 is a flowchart showing an outline of content and annotation information processing in the information relevance calculation unit 110.

  As shown in FIG. 3, the information relevance calculation unit 110 first manages the content set managed by the annotation storage device 100 by the content selection unit 111 that meets the conditions input from the input / output device 20 or the like. (Step S1). The conditions for selecting the content may be arbitrary, such as selection by a menu, selection by creation date and time, selection by a search keyword, and the like.

  Next, the AA relevance calculation unit 112 acquires the annotation group added to the content whose relevance is to be calculated in the selected content set from the annotation storage device 100 (step S2). Subsequently, the AA relevance calculating unit 112 executes a process for calculating the AA relevance for an arbitrary combination of acquired annotations (step S3).

  Next, the A-C relevance calculation unit 113 executes a process of calculating the relevance between each annotation and content (A-C relevance) from the A-A relevance calculated in step S3 (step S4). ). The calculated AC relevance level of the annotation group is output to the input / output device 20 or the like and presented to the user. Alternatively, it is stored in the AC relevance storage device 120 (step S5). Furthermore, the annotation groups that are the targets of calculation of the A-C relevance level may be rearranged in the order of high or low A-C relevance level, and the ordered annotation group may be output. If there are a plurality of selected contents, steps S2 to S5 are repeated for each content.

  FIG. 4 is a diagram illustrating an example of annotation information stored in the annotation storage apparatus 100. As shown in FIG. 4, the annotation storage apparatus 100 stores a content information management table 101 and an annotation information management table 102 that store data about content that is a target of calculation of relevance and annotations added to the content. Has been. The annotation storage apparatus 100 outputs data about the content and the annotation added to the content by designating a search condition for selecting the content.

  As shown in FIG. 4A, the content information management table 101 stores a content ID for identifying content, a content name, content description information, and the like. Further, as shown in FIG. 4B, the annotation information management table 102 stores information such as an annotation ID for identifying an annotation and an annotation name.

  The content information management table 101 and the annotation information management table 102 shown in FIG. 4 are examples, and attributes such as name and description may not be stored, and other attributes may be stored.

  FIG. 5 is a diagram illustrating an example of the annotation-content relationship management table (AC relationship management table) 121 stored in the AC relevance storage device 120. In the A-C relationship management table 121, as shown in FIG. 5, the A-C relevance calculated by the information relevance calculation unit 110 is stored for each combination of each content ID and annotation ID. The The A-C relationship management table 121 shown in FIG. 5 is also an example, and information such as the degree of A-C relevance may be stored in another data format.

  Based on the information stored in the annotation storage device 100 described above, the content selection unit 111 searches the annotation storage device 100 for content for which the degree of association is to be calculated, and sets the content for each corresponding content. The added annotation group is acquired, and the acquired annotation group is transmitted to the AA association degree calculation unit 112. The AA relevance calculating unit 112 calculates an AA relevance that is a relevance between arbitrary annotations added to the same content for each content satisfying the condition.

  FIG. 6 shows an example of the AA relevance level for the annotation added to the content according to the embodiment of the present invention. In the figure, annotation A, annotation B, annotation C, and annotation D represent annotations added to the same content a. The degree of association between annotations is expressed as a relationship between two or more annotations. In the present embodiment, the degree of association between annotations will be described as a relationship between two annotations. In two annotations A and B, the degree of association with annotation B viewed from annotation A and the degree of association with annotation A viewed from annotation B are not necessarily the same. This relationship may be defined as a one-way relationship.

[First Example of AA Relevance]
For example, as a first example of the A-A relevance R (A, B) between annotation A and annotation B added to the same content a, another annotation for a content set to which a certain annotation is added is also added. It may be defined by the inclusion rate of the content set. At this time, the AA relevance R (A, B) is given by the following equation (2).

R (A, B) = {number of contents including both A and B} / {number of contents including B}
... Formula (2)
Here, the number of contents including both A and B is the total number of contents to which both annotation A and annotation B are added in the contents set stored in the annotation storage apparatus 100. The number of contents including B is the total number of contents to which annotation B is added in the contents set stored in the annotation storage apparatus 100.

  Content attached when calculating the AA relevance when there is a large difference in the total number of annotated content by using the method described in the first example of the AA relevance It is possible to define an asymmetric relationship such that the AA relevance level is calculated low for annotations with a large number, and the AA relevance level is calculated high for annotations with a small number of attached contents. .

  FIG. 7 is a diagram illustrating an example of content to which annotations are added. FIG. 8 is a diagram schematically illustrating a method of calculating the AA relevance level. For example, consider content with an annotation as shown in FIG. At this time, the relations R (Enoden, train) and R (train, Enoden) between the two annotations “Enoden” and “train” are calculated as shown in FIG. That is, when the number of contents to which the annotation “train” is added is 100, the number of contents to which “Enoden” is added is 25, and the number of contents to which both “train” and “Enoden” are added is 20, The AA relevance is calculated as follows.

R (Enoden, train) = 20/100 = 0.2
R (train, Enoden) = 20/25 = 0.8
[Second Example of AA Relevance]
As a second example of the AA relevance level R (A, B), the AA relevance level may be defined by the co-occurrence frequency of annotations in the same content. At this time, the AA relevance R (A, B) is given by the following equation (3).

R (A, B)
= {Number of contents including both A and B} / {number of contents including A or B} Expression (3)
Here, the number of contents including both A and B is the total number of contents to which annotation A or annotation B is added in the contents set stored in the annotation storage apparatus 100.

  By using the method described in the second example of the AA relevance level, the AA relevance level does not increase when there is a difference in the number of contents with annotations added. It can be prevented that the AA relevance level is calculated high due to the influence of annotations having a large number of contents or annotations added to only a few contents.

[Third Example of AA Relevance]
Further, as a third example of another AA association degree R (A, B), it may be defined by a chi-square value for the appearance frequency of the annotation. At this time, the AA relevance R (A, B) is given by the following equation (4).

  Here, n is the total number of contents existing in the content set stored in the annotation storage apparatus 100, and [A, B], [￣A, ￣B], [￣A, B], [A, ￣ B] is the number of contents including both annotations A and B, the number of contents not including both annotations A and B, the number of contents including annotation B and not including annotation A, and the number of contents including annotation A and not including annotation B It is. Note that “￣” in “￣A” is a symbol on A (the same applies to B).

  [A], [￣A], [B], and [￣B] are the number of contents including annotation A, the number of contents not including annotation A, the number of contents including annotation B, and the contents not including annotation B, respectively. Is a number. When the number of obtained contents is small, correction such as “Yates correction” as described in Reference Document 1 below may be performed.

[Reference 1]: F. Yates, “Contingency Tables Involving Small Numbers and the X ² Test”, Supplement to the Journal of the Royal Statistical Society, Pages: 217-235, 1934.
Since the statistical information is used by using the method described in the third embodiment of the AA relevance level, AA with high accuracy reflecting whether or not there is a correlation in the tendency that each annotation is added. Relevance can be calculated.

  The A-C relevance calculating unit 113 receives the value of the relevance between the annotations calculated by the AA relevance calculating unit 112 as input, and calculates the A-C relevance that is the relevance between the annotation and the content. Perform the calculation.

  In a system for adding annotations to content, the main attributes and features of the content tend to be expressed by many annotations added to the content. Therefore, in this embodiment, when an annotation added to a certain content has a higher degree of association with a larger number of other annotations added to the same content, the degree of association between the annotation and the content increases. Will be described. That is, an annotation having a high relevance level with any annotation in the annotation group added to the content is calculated to have a high AC relevance level.

[First Example of A-C Relevance]
In the example of the annotation group in the content illustrated in FIG. 6, for example, as a first example of the A-C relevance, the A-C relevance S (a, A) of the annotation A with respect to the content a is the same as the annotation A. You may define by the sum total of AA relevance degree with the other annotation group added to the content. At this time, the AC relevance S (a, A) is given by the following equation (5).

S (a, A) = ΣR (t, A) [where Σ is the sum of t∈T] (5)
T is a set including annotation groups other than the annotation A added to the content a.

  By using the method described in the first example of the A-C relevance level, it is only necessary to obtain the total of the calculated A-A relevance levels. Therefore, it is possible to calculate the A-C relevance level with a low calculation cost. it can.

  FIG. 7 shows an example of the display screen of the content with the annotation added. FIG. 9 shows an example of calculating the A-C relevance from the AA relevance of the annotation added to the content. FIG. 9 shows an AA association degree table whose elements are R (A, B) values calculated by the AA association degree calculation unit 112. For example, the content “Kanagawa Prefecture” for the annotation “travel” is shown. The A-C relevance S (tourist attraction in Kanagawa Prefecture, travel) of “no tourist attraction” is calculated as follows.

S (sightseeing spot, travel of Kanagawa)
= R (Kanagawa Prefecture, Travel) + R (Train, Travel) + R (Enoden, Travel) + R (Enoshima, Travel) + R (Kamakura, Travel)
= 0.1 + 0.3 + 0.01 + 0.01 + 0.03 = 0.45
The AC relevance levels for other annotations are calculated in the same manner. From the A-C relevance level calculated in this way, it can be seen that in the AA relevance level table shown in FIG. 9, the A-C relevance level of the tag “Enoshima” is “3.7”, which is the highest. .

[Second Example of A-C Relevance]
As a second example of the A-C relevance S (a, A), the A-C relevance may be defined by the median AA relevance with other annotation groups. At this time, the AC relevance S (a, A) is given by the following equation (6). Let n be the total number of other annotations added to the same content.

When n is an odd number S (a, A) = R (t ′ _{(n + 1) / 2} , A)
・ When n is an even number: Formula (6)
S (a, A) = (R (t ′ _{n / 2} , A) + R (t ′ _{n / 2 + 1} , A)) / 2
Here, t ′ _i is the i-th annotation when the AA association degrees with all other annotations added to the same content are rearranged in ascending order.

  By using the method described in the second example of the A-C relevance, the A-A relevance includes a very large value or an outlier that is smaller than other values. In addition, it is mitigated by the median value, and it can be prevented that the degree of A-C relevance is unduly high or low.

[Third Example of A-C Relevance]
In addition, as a third example of another AC relevance S (a, A), a random surfer having each annotation as a node and a ratio of AA relevance to other annotation groups as edge strength It may be defined by a model (see Reference 2). At this time, the AC relevance S (a, A) is given by the following equation.

[Reference 2]: S. Brin and L. Page, “The anatomy of a large scale hypertextual web search engine”, In Proceedings of the seventh international conference on World Wide Web, Pages: 107-117, 1998.
In the following expression, t _i is an i-th annotation added to the content a, and T _j is a set including annotations other than the annotation t _j among the annotations added to the content a. To do.
1. A matrix M _a having M _a (i, j) as an (i, j) component is defined.

When i = j: M _a (i, j) = 0
When i ≠ j: M _a (i, j) = R (t _j , t _i ) / ΣR (t _j , t)
[Where Σ is the sum of t∈T _j ]
2. All combinations of eigenvalues of M _a and eigenvector sequences are calculated, and a vector V obtained by normalizing the eigenvectors having the maximum absolute value so as to have a length of 1 is obtained.
3. The value of the i-th row of the vector V is S (a, t _i ).

  In the random surfer model, even if the edge strength is the same, an edge with a node having a higher value is more important. In other words, by using the method described in the third example of the A-C relevance level, it is possible to determine which annotations are considered in consideration of the annotation having a high AA relevance level with many other annotations. It is determined whether to attach importance to the AA relevance level. Accordingly, it is possible to calculate the AC relevance with high accuracy.

[Content ranking processing]
By using content with annotations whose AC relevance has been calculated, the content set of the search results when the user performs a search by specifying the annotations in descending order of the relationship with the specified annotations. It is possible to present them side by side. As a result, the user who has performed the search can find and browse content having more features expressed by the annotation, even in the content set with the same annotation.

  At this time, for example, the search result may be rearranged by using the number of tags with the highest A-C relevance in the content, which is the annotation specified in the search condition.

  FIG. 10 is a diagram illustrating a configuration example of a content management apparatus that rearranges content sets. 10, the annotation storage device 100, the information relevance calculation unit 110, and the A-C relevance storage device 120 correspond to the same reference numerals in FIG.

  The content ranking unit 130 rearranges the content set according to the AC relevance level, and includes a content selection unit 131, an annotation rearrangement unit 132, and a content rearrangement unit 133.

  Hereinafter, a processing procedure executed by the content ranking unit 130 illustrated in FIG. 10 will be described. The content ranking unit 130 determines the order of the contents in the search result list obtained by searching for the contents by the following processing procedure.

  (1) In response to a search request from a user, the content selection unit 131 acquires a content set to which a designated annotation is added from the annotation storage device 100.

  (2) The content selection unit 131 reads the A-C relevance level added to the acquired content from the A-C relevance level storage device 120, and the annotations added to each content and their A-C relevance levels Is transmitted to the annotation rearrangement unit 132. However, if the A-C relevance level of the annotation group has not been calculated, the A-C relevance level is calculated using the processing function of the information relevance level calculation unit 110 described above, and the calculation result is sent to the annotation rearrangement unit 132. introduce.

  (3) For each acquired content, the annotation rearrangement unit 132 rearranges the annotations added to the content in descending order of A-C relevance, and transmits each content and annotation to the content rearrangement unit 133.

  (4) The content rearrangement unit 133 rearranges the content in order from the content in which the designated annotation is added to the upper level after the rearrangement. However, if the position of the specified annotation is the same, it is arranged in order from the content with few added annotations. In addition, when the number of annotations added to the content is the same, the specified annotations are arranged in descending order of the AC relevance value. Further, when the annotations have the same AC relevance value, they are arranged in order from the smallest content ID.

  (5) The content rearrangement unit 133 outputs the result of rearranging the content set to the user.

[Related content acquisition processing]
Furthermore, it is possible to find another content related to a certain content by using the content to which the annotation for which the AC relevance degree is calculated is added. As a result, when a user browses a certain content, the user can automatically obtain information for recommending another content related to the browsed content, and can subsequently browse the related content.

  At this time, for example, the cosine similarity may be used as the similarity of the annotation group attached to the content to calculate the content similarity.

  FIG. 11 is a diagram illustrating a configuration example of a content management apparatus that acquires a related content set. In FIG. 11, the annotation storage device 100, the information relevance calculation unit 110, and the AC relevance storage device 120 correspond to the same reference numerals in FIG.

  The related content acquisition unit 140 acquires content related to the currently focused content by using the content with the annotation with the calculated AC relevance level. The vector creation unit 141, the vector It consists of a similarity calculation unit 142 and a related content presentation unit 143.

  Hereinafter, a processing procedure executed by the related content acquisition unit 140 illustrated in FIG. 11 will be described. The related content acquisition unit 140 selects content related to the content currently focused on in the following processing procedure.

(1) vector generating unit 141, the content set _{C = {C i | i =} 1~n C} ( however, n _C is the total number of contents) for each content C _i in the annotations added to the C _i The set T _i = {T _ij | j = 1 to n _Ti } (where n _Ti is the total number of annotations added to the content) is an element, and the value of each element is the annotation T _ij and the _AC of the content C _i An AC relevance vector V _i having a relevance value is defined and transmitted to the vector similarity calculation unit 142. However, if the A-C relevance level of the annotation group has not been calculated, the A-C relevance level is calculated using the processing function of the information relevance level calculation unit 110 described above, and an A-C relevance level vector is created. , To the vector similarity calculation unit 142.

(2) The vector similarity calculation unit 142 calculates the A-C association degree vector V _x and the A-C association degree vector set V = {V _i | i = 1 to n _C , i ≠ x}. The cosine similarity cos (V _x , V _i ) with the C relevance vector V _i is calculated and transmitted to the related content presentation unit 143.

(3) The related content presentation unit 143 obtains a set of AC relevance vectors V _{i in which} the cosine similarity cos (V _x , V _i ) is equal to or greater than a predetermined threshold value. A set of contents C _i corresponding to the C relevance vector V _i is set as a set of contents related to the contents C _x .

  (4) The related content presentation unit 143 outputs a related content set to the user.

  Although the embodiments of the present invention have been described above, the present invention is not limited to the above-described embodiments, and modifications and applications can be made within the technical scope described in the claims.

DESCRIPTION OF SYMBOLS 10 Content management apparatus 20 Input / output apparatus 100 Annotation accumulation | storage device 101 Content information management table 102 Annotation information management table 110 Information relevance calculation part 111 Content selection part 112 AA relevance degree calculation part 113 AC relevance degree calculation part 120 A -C Relevance Level Storage Device 121 A-C Relationship Management Table 130 Content Ranking Unit 131 Content Selection Unit 132 Annotation Rearrangement Unit 133 Content Rearrangement Unit 140 Related Content Acquisition Unit 141 Vector Creation Unit 142 Vector Similarity Calculation Unit 143 Related Content Presentation Part

Claims (8)

A content management device that stores and manages annotations added to content,
Annotation storage means storing annotations added to the content;
A content selection means for selecting designated content and acquiring an annotation group added to the content from the annotation storage means;
An inter-annotation relevance calculating means for calculating a relevance between annotations that expresses the strength of the relationship between annotations in the annotation group by a numerical value;
An annotation-content relevance calculating means for calculating the relevance between the annotation and the content that expresses the strength of the relationship between each annotation and the content from the relevance between the annotations,
A content management apparatus, comprising: an annotation-content relevance output means for storing or outputting the calculated relevance between the annotation and the content. The inter-annotation relevance calculating means is:
A means for calculating the degree of association based on the content rate of a content set to which another annotation is added to a content set to which a certain annotation is added,
Or a means of calculating the relevance by the co-occurrence frequency in the same content of annotation,
The content management apparatus according to claim 1, further comprising at least one of a degree-of-association calculation unit based on an independence test value with respect to an annotation appearance frequency. The annotation-content relevance calculation means includes:
A means for calculating the degree of association by the sum of the degree of association between annotations with other annotation groups,
Or a means for calculating the degree of association by the median degree of association between annotations with other annotation groups,
3. The method according to claim 1, further comprising at least one of a degree-of-association calculation means based on a random surfer model in which a link strength is a ratio of the degree of association between annotations with other annotation groups. Content management device. Based on the annotation group for which the degree of association between the annotation and the content is calculated and the degree of association between the annotation and the content, the order is sorted in descending order of the degree of association between the annotation and the content. The content management apparatus according to claim 1, further comprising output processing means for outputting the annotation group that has been processed. A content management device that stores and manages annotations added to content,
Annotation storage means storing annotations added to the content;
The degree of association between the annotation and the content calculated by the content selection unit, the inter-annotation relevance calculation unit, and the annotation-content relevance calculation unit according to any one of claims 1 to 4. Annotation-content relevance storage means for storing
A certain content based on the information of the annotation group added to each content stored in the annotation storage means and the information on the relation between the annotation and the content stored in the annotation-content relevance storage means An annotation sorting means for rearranging the order of annotations added to the item in descending order of the degree of association between the annotation and the content,
A content management device comprising content sorting means for rearranging the order of content sets based on annotations added to each content in the content set and the degree of association between the annotations and the content . A content management device that stores and manages annotations added to content,
Annotation storage means storing annotations added to the content;
The degree of association between the annotation and the content calculated by the content selection unit, the inter-annotation relevance calculation unit, and the annotation-content relevance calculation unit according to any one of claims 1 to 4. Annotation-content relevance storage means for storing
A vector creation means for creating an annotation having a value of the degree of association between the annotations of the annotation group added to each content stored in the annotation storage means and the content, and a degree of association vector between the contents;
A vector similarity calculation means for calculating a similarity between the relevance vectors between the annotation created for each content and the content;
A content management apparatus comprising: related content presenting means for acquiring and presenting other content related to a certain content based on information on the similarity between the calculated annotation and the relevance vector between the content . An information relevance calculation method executed by a content management apparatus having an annotation storage unit storing annotations added to content,
A content selection step of selecting the specified content and acquiring the annotation group added to the content from the annotation storage means;
An inter-annotation relevance calculation step for calculating a relevance level between annotations that numerically represents the strength of the relationship between annotations in the annotation group;
An annotation-content relevance calculating step for calculating the relevance between the annotation and the content that expresses the strength of the relationship between each annotation and the content from the relevance between the annotations;
An information relevance calculation method, comprising: executing an annotation-content relevance output step for storing or outputting a relevance between the calculated annotation and the content.   An information relevance calculation program for causing a computer to execute the information relevance calculation method according to claim 7.

Images