KR20170027252A - Cluster method for using broadcast contents and broadcast relational data and user apparatus for performing the method - Google Patents

Abstract

The present invention relates to a clustering method by using broadcast content and broadcast-associated data and a user terminal for performing the method. The cluster method generates a story graph for each of a plurality of scenes for broadcast content by using the broadcast content and broadcast-associated data, and generates a cluster for the scene by using the generated story graph.

Description

CLAIMS AND METHODS FOR CLUSTER METHOD FOR USING BROADCAST CONTENTS AND BROADCAST RELATIONAL DATA AND USER APPARATUS FOR PERFORMING THE METHOD,

The following description relates to a cluster method using broadcast content and broadcast-related data, and to a user terminal performing the method. More specifically, the present invention relates to a cluster method using broadcast contents and broadcast- And a user terminal performing a clustering method and a method for dividing the clusters into story-based clusters.

The growth of international OTT operators such as Netflix, Hulu, and Amazon FireTV, and the proliferation of domestic IPTV and CATV are changing the unidirectional consumption pattern of conventional broadcasting contents. In other words, in the past, if the viewer consumes the content to be transmitted in time according to the time, the latest viewer selectively consumes the broadcast content according to the request. Such changes in consumption patterns are promoting changes in broadcasting services.

Particularly, if a viewer manually waits for a part of broadcast contents to be watched, a part desired by a viewer is watched or watched in a Web service or a VoD service of IPTV, or in the case of some contents, . One of the main underlying technologies for realizing such a service is a division technology for broadcasting contents, and thus manual, semi-automatic, and automatic division technology are proposed, and the divided contents are used as basic unit contents of the service.

In the conventional broadcast content segmentation, the sound information is rapidly changed in a method based on the physical change of the content or is divided into scenes in consideration of the change of the screen. Such conventional techniques are based on changes in physical properties, and therefore, they can be applied to different contents that appear in the same story line, such as contents moving to different places, places where various places or people are involved in one event There is a problem that the scenes can not be connected.

Recently, in order to solve the connection problem between these different scenes, a problem of connection between different scenes is solved by dividing broadcasting contents directly by a person or inspecting an automatically divided contents. However, this method takes a lot of time and money to connect different scenes as a person divides and inspects them directly.

Accordingly, there is a need for a method of clustering scenes of broadcast contents in consideration of not only scenes of broadcast contents but also stories.

The present invention can provide a cluster method for generating a story-based cluster constituting broadcast contents by analyzing video, audio, and associated unstructured data for broadcast contents, and a user terminal performing the method.

The present invention relates to a cluster method for constructing a cluster by layering broadcast contents by constructing a story graph for scenes based on physical changes and measuring consistency between story graphs to perform clusters of story units, A user terminal can be provided.

A clustering method according to an exemplary embodiment of the present invention includes receiving broadcast content and broadcast-related data; Determining a plurality of scenes for the broadcast content using the broadcast content and the broadcast-related data; Generating a story graph for each of the determined plurality of scenes; And generating a cluster for a scene using the generated story graph.

The step of determining a scene according to an exemplary embodiment includes: extracting a shot of the broadcast content; Determining a first scene association between a plurality of first scenes from the extracted shots; Determining a second scene association between a plurality of second scenes extracted from the broadcast association data; And generating a scene in which the first scene association and the second scene association are matched with each other.

The step of extracting shots for broadcast contents according to an exemplary embodiment may extract shots of the broadcast contents based on a degree of similarity between a plurality of frames constituting the broadcast contents.

The step of generating a scene according to an embodiment may generate a scene in which first scene association and second scene association are matched based on the similarity between the plurality of first scenes and the second plurality of scenes.

The step of determining a story graph according to an exemplary embodiment of the present invention includes: extracting keywords from broadcast-related data; And generating a story graph including a node corresponding to the keyword and an edge corresponding to the association of the keyword.

The node and the edge according to an exemplary embodiment may have a weight extracted from broadcast time for the broadcast content.

A story graph according to an exemplary embodiment may be represented by a matrix representing a change in a weight of the edge and a matrix representing a weight change of a node.

The step of creating a cluster according to an embodiment includes the steps of: determining consistency of a story graph of each of the scenes; And combining the story graph of each of the scenes based on the determined consistency.

The step of determining consistency according to one embodiment may determine the consistency of the story graph of each of the scenes based on the size of the subgraph shared by the two story graphs.

A subgraph according to an exemplary embodiment of the present invention refers to an overlapped region in which two story graphs overlap, and the consistency in the overlap region can be determined by the size of the subgraph shared by the two story graphs and the density of the shared subgraph have.

A cluster for a scene according to an exemplary embodiment may consist of a discontinuous scene according to the story graph, and may be expressed in a tree form.

A clustering method according to an exemplary embodiment of the present invention includes receiving broadcast content and broadcast-related data; Extracting a shot for the broadcast content based on a degree of similarity between a plurality of frames constituting the broadcast content; Determining a plurality of scenes according to the broadcast content and the broadcast-related data based on the extracted shot; And generating a cluster for a scene in consideration of consistency of the story graph of each of the scenes.

The determining of the plurality of scenes according to an exemplary embodiment may include generating a plurality of initial scenes from the extracted shots; Determining a first scene association between the plurality of initial scenes; Determining a second scene association between scenes in the broadcast-related data using information on scenes extracted from the broadcast-related data; And generating a scene in which the first scene association is matched with the second scene association.

The step of generating a scene according to an exemplary embodiment may generate a scene in which a first scene association and a second scene association are matched based on a similarity between the plurality of initial scenes and a scene extracted from the broadcast association data.

The step of generating a cluster according to an exemplary embodiment may use a story graph of each of a scene corresponding to a node extracted from the broadcast-related data and an edge corresponding to the association of the keyword.

The node and the edge according to an exemplary embodiment may have a weight extracted from broadcast time for the broadcast content.

A story graph according to an exemplary embodiment may be represented by a matrix representing a change in a weight of the edge and a matrix representing a weight change of a node.

The consistency of the story graphs of each of the scenes according to one embodiment may be determined by the size of the subgraphs shared by the two story graphs.

A subgraph according to an exemplary embodiment of the present invention refers to an overlapped region in which two story graphs overlap, and the consistency in the overlap region can be determined by the size of the subgraph shared by the two story graphs and the density of the shared subgraph have.

The cluster method according to an embodiment of the present invention and the user terminal performing the method can reduce the time and cost for providing a broadcast service on a scene-by-scene basis, The service range can be extended by providing the broadcasting contents of the unit.

FIG. 1 is a diagram illustrating a general configuration of a user terminal for dividing broadcast contents according to an embodiment into clusters in story units.
2 is a diagram for explaining an operation of determining a plurality of scenes for broadcast contents according to an exemplary embodiment.
3 is a diagram illustrating a storage format of a plurality of scenes for broadcast contents according to an exemplary embodiment.
4 is a flowchart illustrating a procedure for extracting a story graph of broadcast content according to an exemplary embodiment of the present invention.
FIG. 5 is a view showing the shape of a story graph of each of scenes according to an embodiment.
6 is a flowchart illustrating a procedure for creating a cluster for a scene according to an exemplary embodiment of the present invention.
7 is a flowchart illustrating a cluster method according to an embodiment.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

FIG. 1 is a diagram illustrating a general configuration of a user terminal for dividing broadcast contents according to an embodiment into clusters in story units.

Referring to FIG. 1, the user terminal 100 determines a plurality of scenes for the broadcast content using broadcast content and broadcast-related data, and generates a cluster for a scene using a story graph for each of the determined plurality of scenes. Lt; / RTI > Here, the user terminal 100 may be a device for displaying broadcast content on the screen of the user terminal 100. [ Alternatively, the user terminal 100 may be a device for receiving broadcast content from outside and providing broadcast content received by a separate display device. In addition, the user terminal 100 may include an apparatus for collecting, processing, analyzing and extracting semantic clusters by inputting broadcast contents. In one example, the user terminal 100 may include a display, such as a television, set-top box, desk top, or other device capable of displaying broadcast content via a separate device.

The user terminal 100 may include an image-based shot extracting unit 110, a shot-based scene extracting unit 120, a story graph extracting unit 130, and a cluster generating unit 140.

The image-based shot extracting unit 110 may receive the broadcast content 210 and the broadcast-related data 220. The image-based shot extracting unit 110 may extract a shot of the broadcast content 210 based on the inter-frame similarity. The similarity between frames can represent a result calculated based on the difference in color, texture, area, etc. of the object or background constituting the frame. For example, the similarity between frames can be calculated using cosine similarity, euclidean distance, or the like based on a feature vector such as a color histogram and motion extracted from a "frame ".

 As a result, the image-based shot extracting unit 110 extracts shots of the broadcast contents 210 based on the degree of similarity between the frames, and the broadcast contents can be represented by a sequence of shots.

The broadcast-related data 220 may include information such as subtitles, scripts, and the like associated with the broadcast content 210. The image-based shot extracting unit 110 may extract shots of broadcast contents based on the degree of similarity between a plurality of frames constituting broadcast contents.

Specifically, the image-based shot extracting unit 110 may extract shots of the broadcast contents in consideration of a physical change of the broadcast contents 210. To this end, the image-based shot extracting unit 110 may extract a voice feature and an image feature from the broadcast content 210. [ The image-based shot extracting unit 110 may extract shots of broadcast contents according to physical changes based on the extracted image features.

The shot-based scene extracting unit 120 may determine a plurality of scenes for the broadcast content using the broadcast content and the broadcast-related data. The shot-based scene extracting unit 120 can determine a plurality of scenes for the broadcast contents based on the temporal relevance of the extracted shots and the information about scenes extracted from the broadcast-related data.

Specifically, the shot-based scene extracting unit 120 may determine a first scene-relevance among the plurality of first scenes from the extracted shot. Here, the first scenes may refer to a plurality of initial scenes from a shot, and the shot-based scene extractor 120 may determine a first scene correlation between a plurality of initial scenes. That is, the first scene association may indicate the association between the shots of the broadcast contents.

The shot-based scene extracting unit 120 may determine a second scene association between a plurality of second scenes extracted from the broadcast-related data. Here, the second scenes may be information on a scene extracted from the broadcast-related data, and the shot-based scene extracting unit 120 may extract second scene- Can be determined. The shot-based scene extracting unit 120 can determine a plurality of scenes for the broadcast content by generating scenes matching the first scene association and the second scene association. Here, the shot-based scene extracting unit 120 generates a scene in which the first scene association and the second scene association are matched to each other as much as possible, and assumes that there exist a plurality of pieces of data indicating association between the broadcast content and the broadcast- , A scene that is matched to the maximum may mean a scene in which a first scene-relevance and a second scene-relevance match most closely among a plurality of data.

The story graph extracting unit 130 may generate a story graph for each of the determined plurality of scenes. Specifically, the story graph extracting unit 130 can extract keywords from the broadcast-related data. The story graph extracting unit 130 can generate a story graph composed of nodes corresponding to the keywords and edges corresponding to the associations of the keywords. Here, the node and the edge may represent a weight extracted from the broadcast time of the broadcast content, and the story graph may be expressed by a matrix representing a weight change of the edge and a matrix representing a weight change of a node.

The cluster generating unit 140 may generate a cluster for a scene using the generated story graph. Here, the cluster generating unit 140 can generate clusters for scenes in consideration of semantic consistency of story graphs. Clusters for scenes are non-contiguous scenes according to story graphs, and are multi-layered semantic clusters. Lt; / RTI >

As a result, the cluster method proposed in the present invention receives broadcast contents 210 and broadcast related data 220 and generates a story-based semantic cluster based on the received broadcast contents 210 and broadcast related data 220 can do. The semantic cluster of the story unit generated through the cluster method can be stored and stored in the cluster storage 150.

The cluster method proposed by the present invention proposes a story-based segmentation technique for broadcast contents. The segmentation of a story unit proposed in the present invention means dividing a scene so as to show a plurality of story lines constituting broadcast contents . To this end, the cluster method proposed in the present invention generates a story graph representing a story of a scene from a scene divided on the basis of a shot extracted based on the similarity between frames according to broadcast contents, and generates a semantic consistency The scenes can be layered and combined. In the present invention, the broadcast content finally divided into story units can be expressed as " semantic cluster ".

2 is a diagram for explaining an operation of determining a plurality of scenes for broadcast contents according to an exemplary embodiment.

Referring to FIG. 2, the shot-based scene extracting unit 120 may determine a plurality of scenes for the broadcast content using broadcast content and broadcast-related data. Specifically, the shot-based scene extracting unit 120 may extract a relevance to a scene corresponding to each of the broadcast content and the broadcast-related data, and may determine a plurality of scenes for the broadcast content according to the extracted relevance.

(1) broadcast contents

The shot-based scene extracting unit 120 may extract the relevance to the scene based on the broadcast contents. In detail, the shot-based scene extracting unit 120 may determine a first scene-relevance among the plurality of first scenes from the shots extracted from the image-based shot extracting unit 110. [ Here, the shot-based scene extracting unit 120 may generate an initial scene based on the similarity between the shots of the broadcast contents. In this case, the initial scene may mean a scene used for determining the first scene association.

The shot-based scene extracting unit 120 may determine a first scene association between a plurality of generated initial scenes. That is, the shot-based scene extracting unit 120 may calculate the relevance between the scenes constructed through the inter-scene correlation measurement. The shot-based scene extracting unit 120 may extract a video feature, a voice feature, and the like of the broadcast content corresponding to the shot section after extracting the shot. The shot-based scene extracting unit 120 may measure the relevance between shots by comparing extracted feature vectors using a conventional vector similarity calculation technique (cosine similarity, euclidean distance, etc.).

(2) broadcast-related data

The shot-based scene extracting unit 120 may analyze the broadcast-related data to determine a second scene association between a plurality of second scenes. In detail, the shot-based scene extracting unit 120 extracts information related to a plurality of scenes from the broadcast-related data, and based on the extracted information, 2 Scene associations can be extracted. The shot-based scene extracting unit 120 may extract the relevance of scenes existing in the association data by analyzing the broadcast-related data (script, subtitles). For example, the shot-based scene extracting unit 120 extracts the subtitles present in the scenes in the case of the subtitles and extracts the scenes on the scenes in the case of the scenes to extract the relevance information between the scenes constituting the broadcast contents .

Thereafter, the shot-based scene extracting unit 120 may generate a scene in which the first scene association and the second scene association match. Specifically, the shot-based scene extracting unit 120 may generate scenes in which the first scene association and the second scene association match based on the similarity between the plurality of first scenes and the second plurality of scenes. That is, the shot-based scene extracting unit 120 extracts 1) the direct similarity between the first scenes extracted from the broadcast contents and the second scenes extracted from the broadcast-related data, and 2) the similarity between the first scenes and the second scenes A plurality of scenes for the broadcast content can be determined so that the association of the broadcast contents is matched.

As a result, the shot-based scene extracting unit 120 may construct a plurality of scenes for the broadcast contents through the relevance matching, scenes of the broadcast contents, and information about the scenes obtained from the broadcast-related data. The information on the scene is information used for the association matching, and may include information on the first scene association and the second scene association.

3 is a diagram illustrating a storage format of a plurality of scenes for broadcast contents according to an exemplary embodiment.

Referring to FIG. 3, the shot-based scene extracting unit 120 extracts a shot based on a similarity between a plurality of first scenes and the second plurality of scenes, A plurality of scenes can be generated. The shot-based scene extracting unit 120 may display a data structure for storing a plurality of scenes for the generated broadcast content.

Specifically, the broadcast content is a set of a plurality of scenes, which is expressed as C = {S ₁ , S ₂ , S ₃ , ..., S _m }, _Si denotes an i th shot, And may have a number ( _Bi ) and an end frame number ( _Ei ). In addition, each of the plurality of scenes may be a set of one or more frames, and one scene constituting broadcast contents includes a start frame and a last frame, and may include a scene feature vector and a speech feature vector have. One scene constituting broadcast contents has association data linked to the scene, and association data can be composed of keywords.

In addition, the association data may be implemented by converting a graph, a tree, or the like expressing the relationship between the keywords to express the keyword extracted through the broadcast-related data. Here, the associated data can be utilized as information for being converted into a story graph according to the extracted scene.

4 is a flowchart illustrating a procedure for extracting a story graph of broadcast content according to an exemplary embodiment of the present invention.

Referring to FIG. 4, the story graph extracting unit 130 may generate a story graph for each of a plurality of determined scenes. Specifically, the story graph extracting unit 130 can extract keywords from the broadcast-related data. Here, the keyword extracted from the broadcast-related data can be implemented as association data, and it can be utilized as information to be converted into a story graph according to the extracted scene.

In other words, the association data including the keyword extracted from the broadcast association data can be converted into a story graph for each scene. That is, the story graph extracting unit 130 can generate a story graph composed of nodes corresponding to the keywords and edges corresponding to the associations of the keywords. Story graphs can be defined as 1) nodes, edges and nodes, and 2) weights for nodes and edges.

A node is a keyword extracted from association data, and an edge may mean association between keywords. The node and the edge may include a weight extracted from the broadcast time for the broadcast content. Therefore, a story graph composed of nodes and edges suggested by the present invention can be represented by an NxN matrix. Here, N is the number of nodes, and the value of the matrix may be a numerical value of the relevance of the edge.

In other words, the story graph extracting unit 130 may represent a matrix representing a weight change of an edge and a matrix representing a weight change of a node, and the matrix expressed thereby may have the form shown in FIG. 5, Stored and managed. The detailed configuration will be described with reference to FIG.

FIG. 5 is a view showing the shape of a story graph of each of scenes according to an embodiment.

Referring to FIG. 5, the story graph extracting unit 130 may perform a node building function and an edge building function based on information about a node and an edge. In addition, the story graph extracting unit 130 may add a weight according to time t to each node and an edge, including a node building function and an edge building function.

In other words, the story graph extracting unit 130 may add a weight according to time t to each node and an edge of the story graph considering the temporal flow of the scene. Therefore, the story graph can be defined as an N x N x T matrix (Fig. 5 (a)) for accommodating changes in the weight of edges and an N x T matrix (Fig. 5 have.

In addition, the story graph extracting unit 130 may calculate the weights according to time with a survival function, a forgetting curve, or the like in order to add weight to each node and edge over time.

6 is a flowchart illustrating a procedure for creating a cluster for a scene according to an exemplary embodiment of the present invention.

Referring to FIG. 6, the cluster generator 140 may perform a function of combining scenes by measuring consistency based on the generated story graph. In other words, the cluster generating unit 140 can generate a cluster for scenes using the generated story graph. To this end, the cluster generator 140 may repeatedly perform the function of combining the story graph and the function of measuring the story consistency to generate the semantic cluster.

Specifically, the cluster generating unit 140 can determine the consistency of the story graph of each of the scenes. Here, the cluster generating unit 140 can determine the consistency of the story graph of each scene based on the size of the subgraph shared by the two story graphs. In this case, the consistency for combining the story graphs may mean a result obtained from the measured values by measuring the degree of stacking of the story graphs. That is, the consistency for combining the story graphs may be a value measured based on the size of the subgraphs shared by the two graphs.

Here, the subgraph means one area that overlaps most in the overlap of the story graph, and the story consistency of the area can be calculated through the overlap of the graph and the density of the overlapped subgraph. Size refers to a shared entity between clusters based on two story graphs, and density can refer to the relationship between shared entities. That is, story consistency may mean measuring the degree to which the same entity (e.g., person, place, event, etc.) has the same relationship.

The cluster generating unit 140 selects one of the story graphs having the greatest story consistency among all the story graphs generated for each of the determined plurality of scenes and repeats the process of combining the selected story graphs repeatedly until one top- . ≪ / RTI > Thus, one broadcast content is represented by a semantic cluster tree, each node of the tree contains a linked story, and the story can be expressed in the form of a combined graph. As a result, if the broadcast content is composed of one semantic cluster tree using the semantic cluster, the result can be stored in the semantic cluster storage 150.

7 is a flowchart illustrating a cluster method according to an embodiment.

In step 701, the user terminal may receive broadcast content.

In step 702, the user terminal may receive broadcast related data.

In step 703, the user terminal can extract a voice feature for a scene based on the input broadcast content.

In step 704, the user terminal extracts video features of the scene based on the input broadcast content, and extracts shots of broadcast content based on the extracted video features in step 706. In other words, the user terminal can extract shots of the broadcast contents considering the physical change of the broadcast contents 210. The user terminal may then determine a first scene association between a plurality of first scenes from the extracted shot

In step 705, the user terminal extracts a keyword from the broadcast-related data, and determines a second scene association between a plurality of second scenes extracted based on the extracted keyword in step 707.

In step 708, the user terminal can determine a plurality of scenes by generating scenes matching the first scene associativity and the second scene associativity. In other words, the user terminal can determine a plurality of scenes for the broadcast content based on the second scene association extracted from the voice feature, the first scene association, and the broadcast association data extracted from the broadcast content.

In step 709, the user terminal may generate a story graph for each of the determined plurality of scenes. In other words, the user terminal can extract a keyword from the broadcast-related data, and generate a story graph composed of a node corresponding to the extracted keyword and an edge corresponding to the association of the keyword.

In step 710, the user terminal may generate a cluster for the scene using the generated story graph.

The methods according to embodiments of the present invention may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions recorded on the medium may be those specially designed and constructed for the present invention or may be available to those skilled in the art of computer software.

While the invention has been shown and described with reference to certain preferred embodiments thereof, it will be understood by those of ordinary skill in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims. This is possible.

Therefore, the scope of the present invention should not be limited to the described embodiments, but should be determined by the equivalents of the claims, as well as the claims.

100: user terminal
110: Image-based shot extraction
120: Shot-based scene extraction
130: Story Graph Extraction
140: Creating a cluster
150: Cluster storage
210: broadcast contents
220: broadcast association data

Claims (20)

Receiving broadcast content and broadcast-related data;
Determining a plurality of scenes for the broadcast content using the broadcast content and the broadcast-related data;
Generating a story graph for each of the determined plurality of scenes; And
Generating a cluster for a scene using the generated story graph;
≪ / RTI > The method according to claim 1,
Wherein the step of determining the scene comprises:
Extracting shots of the broadcast contents;
Determining a first scene association between a plurality of first scenes from the extracted shots;
Determining a second scene association between a plurality of second scenes extracted from the broadcast association data; And
Generating a scene in which the first scene association and the second scene association are matched
≪ / RTI > 3. The method of claim 2,
Wherein the step of extracting a shot for the broadcast content comprises:
And extracting shots of the broadcast contents based on a degree of similarity between a plurality of frames constituting the broadcast contents. 3. The method of claim 2,
Wherein the generating the scene comprises:
And generating a scene in which first scene association and second scene association are matched based on a similarity between the plurality of first scenes and the second plurality of scenes. The method according to claim 1,
The step of determining the story graph
Extracting keywords from broadcast-related data; And
Generating a story graph composed of a node corresponding to the keyword and an edge corresponding to the association of the keyword
≪ / RTI > 6. The method of claim 5,
The node and the edge,
And a weight extracted from a broadcast time of the broadcast content. The method according to claim 6,
In the story graph,
And a matrix representing a weight change of the edge and a matrix representing a weight change of a node. The method according to claim 1,
Wherein the generating the cluster comprises:
Determining consistency of the story graph of each of the scenes; And
Combining the story graphs of each of the scenes based on the determined consistency
≪ / RTI > 9. The method of claim 8,
Wherein determining the consistency comprises:
A clustering method that determines the consistency of each story graph based on the size of subgraphs shared by two story graphs. 10. The method of claim 9,
In the subgraph,
Means an overlap area in which two story graphs overlap,
The consistency in the overlapping area
The clustering method is determined by the size of the subgraph shared by the two story graphs and the density of the shared subgraph. The method according to claim 1,
Wherein the cluster for the scene comprises:
And a non-contiguous scene according to the story graph, the clustering method being expressed in a tree form. Receiving broadcast content and broadcast-related data;
Extracting a shot for the broadcast content based on a degree of similarity between a plurality of frames constituting the broadcast content;
Determining a plurality of scenes according to the broadcast content and the broadcast-related data based on the extracted shot; And
Generating a cluster for a scene in consideration of consistency of a story graph of each of the scenes
≪ / RTI > 13. The method of claim 12,
Wherein the determining of the plurality of scenes comprises:
Generating a plurality of initial scenes from the extracted shots;
Determining a first scene association between the plurality of initial scenes;
Determining a second scene association between scenes in the broadcast-related data using information on scenes extracted from the broadcast-related data; And
Generating a scene in which the first scene association and the second scene association are matched
≪ / RTI > 14. The method of claim 13,
Wherein the generating the scene comprises:
And generating a scene in which the first scene association and the second scene association match based on the similarity between the plurality of initial scenes and the scene extracted from the broadcast association data. 13. The method of claim 12,
Wherein the generating the cluster comprises:
A story graph of each of a scene corresponding to a keyword extracted from the broadcast-related data and an edge corresponding to an association of the keyword. 16. The method of claim 15,
The node and the edge,
And a weight extracted from a broadcast time of the broadcast content. 17. The method of claim 16,
In the story graph,
And a matrix representing a weight change of the edge and a matrix representing a weight change of a node. 13. The method of claim 12,
The consistency of each of the scenes &
The clustering method is determined by the size of the subgraph shared by two story graphs. 19. The method of claim 18,
In the subgraph,
Means an overlap area in which two story graphs overlap,
The consistency in the overlapping area
The clustering method is determined by the size of the subgraph shared by the two story graphs and the density of the shared subgraph. A computer-readable recording medium on which a program for executing the method according to any one of claims 1 to 19 is recorded.

Images