JP4917412B2 - Content extraction apparatus, method, program, and recording medium - Google Patents

Description

  The present invention relates to a content extraction technique, and in particular, a content extraction device, a content extraction method, and a program for the content extraction method for extracting and displaying content from a recording medium that receives and records a digital broadcast signal such as a television device. And a recording medium on which the program is recorded.

  Conventionally, processing such as fast-forward playback has been performed in order to find content that the user wants to see from content recorded with broadcast video. However, as the capacity of storage media increases, searching for the desired content by fast-forward playback from all the recorded content is time consuming and laborious.

  Therefore, recently, a method of displaying a list in a directory format for recorded contents has been performed. However, when the amount of information becomes large, the hierarchical structure becomes deep in the directory format classification, and it takes time to search for the desired content.

  On the other hand, a search method using natural language based on text information given to content has also been proposed. However, this kind of search method is efficient when the content you want to see is clear, but should be used when the user does not have a clear intention or when you cannot express your interests in words. Is difficult.

  In view of this, a method has been proposed in which information is visually classified and arranged so as to be visible to the user in order to make it possible to search for information even when the user does not have a clear intention. For example, in the following Non-Patent Document 1, when a content is a text document, a keyword vector (meaning vector) is generated by extracting a keyword in the text document and an appearance frequency of the keyword, A method is described in which the content is two-dimensionally arranged by applying the method, and the user browses the completed content map.

  Non-Patent Document 2 below proposes a specific method for realizing a spatial arrangement of contents. Based on the concept base (collection of semantic vectors), word information of classification categories represented by predetermined characteristic words is arranged two-dimensionally using multidimensional scaling (this is referred to as “reference map”) And a method for arranging the contents there. In this case, meta information such as text such as a classification category and explanatory text or keywords is assigned to the content in advance, and the similarity (distance) between the meta information and the concept base or reference map is calculated, A semantic vector is calculated, and content is classified and arranged on a reference map using a multidimensional scaling method.

  In the following Patent Document 1, on the assumption that a classification category is assigned to content (image, video, music, text information, etc.), the content to which the classification category information is assigned is reflected in the classification structure. A method is described in which contents are clustered for each classification category by mapping, and the structure of the classification system is reflected in the map.

  FIG. 22 is a diagram showing a configuration example for explaining the concept of the prior art for classifying and arranging contents, and FIG. 23 is a flowchart showing the processing flow. First, the concept vector calculation processing unit 107 illustrated in FIG. 22 reads the outline explanation of the content stored in the content DB 115 (step S501), and calculates the concept vector. At this time, the concept vector is represented as a multidimensional vector. The calculated concept vector is stored in the content DB 115 in association with each content (step S502). Next, the two-dimensional coordinate calculation processing unit 105 calculates the distance between the contents for all the two content combinations using the concept vector of each content (step S503). Then, the position information of the content on the two-dimensional plane is calculated by the multidimensional scale construction method (step S504). The multidimensional scaling method is a compression algorithm from a high-dimensional vector space to a low-dimensional space. A map image is created based on the position information thus calculated (step S505). The content display unit 110 displays the created map image of the content in response to a user request input from the input unit 117 (step S506). Non-patent document 3 can be referred to for a method of calculating a concept vector from the summary explanatory text.

  In the method of spatially arranging the contents represented by Non-Patent Documents 1 and 2 and Patent Document 1, content with a small difference in similarity gathers nearby, and content with a large difference in similarity gathers far away. It is possible to arrange based on the similarity between contents. That is, it becomes easy to grasp the tendency (distribution state) of the recorded content as a whole and the interrelationship of the content, and the content can be freely walked on the content map as shown in the content display unit 110 of FIG. This makes it possible to select the content, thereby improving the convenience of content selection. In particular, since the above method does not have a clear selection intention for content selection, it is not suitable for content selection when the content genre or content selection method using individual metadata or keywords added to the content is not suitable. It is effective as a content display method.

James A. Wise, et al. “Visualizing the non-visual: Spatial analysis and interaction with information from text documents”, Proc. Of IEEE information on Visualization, '95, pp. 51-58. (1995) Goro Fujita “Contents Visualization Based on Integrated Use of Classification Information and Language Information” DEWS2003 1-P-03, Mar. (2003) Kumamoto, et al., “Application to Concept-Based Information Retrieval—Characteristic Evaluation of Concept-Based Retrieval—” Science Technique AI 98-63 (1999). JP 2003-323454 A

  However, as described in Non-Patent Documents 1 and 2 and Patent Document 1, a content map based on the similarity between contents is displayed to the user, and the user is allowed to select content from the content map based on the similarity. May cause the following problems.

1) When the degree of similarity between recorded contents is generally poor, as shown in FIG. 24, the distribution 121 due to the spatial arrangement of contents displayed on the display unit 110 spreads thinly and widely, and is a clue for selecting viewing candidates. I cannot find a content group.
2) Or, as shown in FIG. 25, a large number of content groups 121a composed of a small number of contents are generated.
3) When there is a bias in the similarity of recorded content, such as when only content with high similarity is recorded, the spatial arrangement of the content is biased, and as shown in FIG. As a result, a very small number of content groups 123 including a large number of contents are generated. That is, the number of content groups serving as viewing candidates is too small and the number of contents included in one content group is too large.

When the above phenomenon occurs, the following problems occur in the conventional method.
1) When a content group does not occur as shown in FIG. 24, there is no clue to selecting a content, and it is difficult to grasp what kind of content is present in what location.
2) When the number of content groups becomes too large as shown in FIG. 25, it is difficult to select which content group is to be checked as a clue to selecting content, and it takes time to check the contents of the content group. It takes.
3) When the number of content groups is reduced as shown in FIG. 26, it becomes easy to select a target content group as a clue to selecting content, but on the other hand, the check is included in each content group. Too much content has to be done.

  In view of the above problems, the present invention aims to provide a content extraction technique that facilitates selection of viewing candidates even when the similarity between recorded contents is poor or even when the similarity is biased. Yes.

  According to one aspect of the present invention, a content extraction apparatus that searches for content from a plurality of contents having arrangement information in a content space based on the similarity between contents, and has a predetermined similarity with a certain content A plurality of contents extracted based on a distance in the content space having a correspondence relationship with the similarity, and a content extracting unit capable of extracting a certain number (n) of contents, and extracted by the content extracting unit There is provided a content extraction device including a control unit that performs control to display content on a display unit.

  According to another aspect of the present invention, there is provided a content extraction method for searching for content from a plurality of contents having arrangement information in the content space based on the similarity between the contents. Extracting a certain number (n pieces) of content based on the distance in the content space having a correspondence relationship with the similarity, and displaying the extracted content. There is provided a content extraction method characterized by comprising:

  The present invention includes a program for causing a computer to execute the steps described above, or a computer-readable recording medium for recording the program.

By extracting a certain number of viewing candidates from a content map created based on the similarity between content according to certain rules,
(1) In order to achieve this object, the first configuration of the content extraction apparatus of the present invention includes a content DB, a concept vector calculation processing unit, and an n (n is a natural number of 2 or more) dimensional coordinate calculation processing unit. A content placement coordinate calculation unit, a behavior detection unit that detects a user's behavior, and a content extraction unit that extracts content that is a viewing candidate based on the content placement coordinates calculated by the n-dimensional coordinate calculation processing unit. ing.

  Then, the content extraction unit is located within a range of a predetermined distance r ± Δ1 (where Δ1 is 0 <Δ1 ≦ r / 2) from the content Ki serving as a content selection reference on the n-dimensional arrangement coordinates. A content candidate acquisition unit that acquires content to be viewed from the content DB as viewing candidate content, and whether or not the number N of viewing candidate content acquired by the arrangement coordinate acquisition unit is greater than a preset number n When the content candidate confirmation unit has a content number N larger than the predetermined number n, the region of distance r ± Δ1 is divided into n in order to narrow it down to n, the centroid position of the content distribution of each divided region is obtained, and the centroid A content narrowing unit that extracts content existing at a position closest to the position.

(2) According to a second configuration of the content extraction device of the present invention, when the configuration of the content narrowing unit has a content number N larger than a predetermined number n, the content extraction unit has a distance r ± Δ1 in order to narrow it down to n. This is a content narrowing-down unit that extracts the top n contents of evaluation values based on a preset content evaluation criterion from the included contents, and is different from the first configuration of the content extraction apparatus of the present invention.

  In the present invention, the user can select content from content candidates extracted and displayed based on the similarity between the contents, and there is an advantage that selection of viewing candidates is facilitated.

  As described above, the present invention is based on a technology (Non-Patent Document 1) that arranges contents in a space based on the similarity between the contents. For each content, a multi-dimensional vector (semantic vector) is created in advance using characteristic vocabulary (such as EPG information), video information (such as color information), and acoustic information (such as volume), and the similarity between the contents Is calculated and the content is arranged in a multidimensional space. Or you may arrange | position in two-dimensional space by arrangement | positioning by normal distribution, or a multidimensional scale structure.

  As a result, in the content space, content with high similarity is arranged near and content with low is arranged far away. However, it is not necessary to actually create a map, and it is only necessary to have arrangement coordinates. The content map can be multi-dimensional, but in the following description, a case where content is arranged in a two-dimensional space will be described as an example. Further, a case will be described where the extracted content is displayed around the content that was viewed immediately before the search (20 in FIG. 1).

  Hereinafter, a content extraction technique according to an embodiment of the present invention will be described with reference to the drawings.

First, the content extraction technique according to the first embodiment of the present invention will be described.
The present embodiment is an embodiment corresponding to the solution (1). This embodiment is one of the methods for extracting n representative contents from a predetermined number n (for example, 8) of contents having the same similarity as a certain content Ki, The circumference of a predetermined distance r from a certain content Ki is divided into a predetermined number n (eight in this case), the center of gravity of each region is calculated, and the content existing at the center of gravity is extracted and displayed. It is characterized by that.

  As shown in FIG. 1, the content extraction apparatus A according to the present embodiment includes a behavior detection unit 21 that detects a user's behavior in addition to the content arrangement coordinate calculation unit 6 having the same configuration as the conventional technology (FIG. 22). And a content extraction unit 7 for extracting content that is a viewing candidate based on the content arrangement coordinates.

  The prior art (FIG. 22) is characterized by creating a content map, but in the present embodiment, it is not always necessary to create a content map. Further, the arrangement coordinates of the content may be multi-dimensional instead of two-dimensional. However, in the following description, a case where two-dimensional arrangement coordinates are used is taken as an example. FIG. 1 is a functional block diagram showing a configuration example of the content extraction apparatus A according to the first embodiment of the present invention. As shown in FIG. 1, the content extraction apparatus A according to the present embodiment includes the following components.

1) Content arrangement coordinate calculation unit 6: It has the same configuration as the prior art (FIG. 22), and includes a content DB 2, a concept vector calculation processing unit 6c, a two-dimensional coordinate calculation processing unit 6b, and a map creation processing unit 6a. Yes. However, the map creation processing unit 6a is arbitrarily provided. In the content DB 2, content arrangement coordinates are recorded together with the content.
2) Behavior detection unit 21: For example, it is configured to include sensors such as a user's remote control operation, face orientation, line of sight, body temperature, heart rate, etc., and always detects what content the user is viewing and searches for the user Detect start timing.
3) Content extraction unit 7: An arrangement coordinate acquisition unit 7a, a candidate confirmation unit 7b, and a gravity center narrowing unit 7c are provided.
3-1) The arrangement coordinate acquisition unit 7a acquires the arrangement coordinates of the content located at a predetermined distance r from the content DB 2b based on the content Ki that was viewed immediately before the start of the search.
3-2) The candidate confirmation unit 7b determines whether or not the acquired content number N is greater than the predetermined number n.
3-3) When the content number N is larger than the predetermined number n, the center-of-gravity narrowing unit 7c divides the circumferential region of the distance r into n in order to narrow down to n, and obtains the center-of-gravity position of each region. Extract content located at the center of gravity.

  Next, a method of extracting a predetermined number n contents (for example, 1 to 8 contents that can be displayed as a list on the display) as viewing candidates in the center-of-gravity narrowing unit 7c based on the similarity between contents will be described. The content displayed on the display unit may be the content itself, or any content that actually displays the content such as a thumbnail thereof.

  First, when the number N of contents having a predetermined similarity with a certain content Ki (for example, the content viewed immediately before the search) is less than the predetermined number n, all contents are extracted and displayed. When N is larger than n, it is necessary to narrow down to n. In this solution, the center of gravity of the content group is used as means for extracting n representative contents based on the similarity between contents.

  Next, a method using the center of gravity of the content group will be described. The content space is divided into n (for example, 1 to 8) areas based on the degree of similarity with a certain content Ki (for example, content that was viewed immediately before the search). Next, the barycentric position of each region is obtained, content (n pieces in total) located there is extracted as representative content, and content having the arrangement coordinates is extracted from the content DB 2a and displayed. Although not shown in the present embodiment, the content DB 2a itself may be stored on the network.

(Division method into n regions)
FIG. 3 is a diagram illustrating a display example of the content space. A large number of contents 41 are displayed on the content display unit 20.

(1) Method of extracting n pieces having the same degree of similarity as the content Ki:
(A) In order to extract n pieces having the same degree of similarity as the content Ki, as shown in FIG. 4, n pieces of the circumference 43 with a predetermined distance r (L) centered on the content Ki (corresponding to (1)) (L To). FIG. 4 shows an example of equally dividing into n pieces. The equally divided area 43 is an area from 2) to 9). The center-of-gravity content of the content group located on the circumference of each region 43 is extracted. By selecting content on the same circumference, eight types of content having the same similarity are selected.
FIG. 5 is a diagram illustrating an example in which the circumference L is unequally divided into n pieces. Since there are a plurality of contents on the circumference 43 of the distance r, the center of gravity of each of the n-divided areas is obtained, and a total of n contents are extracted one by one. For example, the content Ki (43) currently being viewed is displayed at the center of the screen 20 in FIG. 9 and the contents (n total: 43-1 to 43-8) positioned at the center of gravity of each region are displayed around the user. (FIG. 9). As a result, the user can select content from n representative contents having a certain degree of similarity with the content Ki.
(A) In (A) above, when the circumference of the distance r is divided, it may be as follows instead of dividing it into n equal parts. That is, all the contents 41 are previously clustered by the K-means method or the like based on the Euclidean distance between the semantic vectors of the contents 41 and divided for each cluster (FIG. 6: 45-1 to 45-4). .
(C) In the above (a), n pieces may be extracted randomly one by one in the divided n areas.

(2) Method of extracting n contents having different degrees of similarity from the contents Ki:
(A) To extract n pieces of content having a degree of similarity different from that of the content Ki, the content Ki (1) is divided into n parts concentrically (FIG. 7), and the contents are randomly selected one by one from each divided region. Are extracted (41a, 41b, 41c,...). As a result, the user can select n representative contents having different degrees of similarity from the contents Ki. In the figure, regions are concentrically separated from the current content at constant distances, and content is randomly extracted from each donut-shaped region. Thereby, eight types of contents having different degrees of similarity can be selected.
(A) After concentrically dividing in (a) above, the circumference is further divided into m by L around the content Ki (FIG. 8). Then, the center of gravity of each n × m divided area (for example, A1) is obtained, and the contents (total n pieces) of the center of gravity position are extracted in each of the m divided areas. Even in this method, the user can select from eight types of content Ki having different degrees of similarity with the content Ki currently or immediately before viewing.

(Basic configuration example of the device)
FIG. 1 is a functional block diagram showing a configuration example of the content extraction apparatus A according to this embodiment. The content extraction apparatus A shown in FIG. 1 receives a digital broadcast signal and records content (television program or the like), and the content DB2 calculates the similarity (distance) between the content and calculates the two-dimensional arrangement coordinates. The arrangement coordinate calculation part 6, the content extraction part 7 which extracts the predetermined number of content to display, and the content display part 20 which displays the extracted content are provided. In the content DB 2, content arrangement coordinates 2b are recorded together with the content 2a. As described above, the behavior detection unit 21, the arrangement coordinate acquisition unit 7a, the candidate confirmation unit 7b, and the gravity center narrowing unit 7c are provided in association with the content extraction unit 7 that is a feature of the present embodiment. The behavior detection unit 21 is, for example, a sensor that detects a user's remote control operation, face orientation, line of sight, body temperature, heart rate, and the like. The behavior detection unit 21 constantly detects which content the user is viewing and starts a user search. To detect. The arrangement coordinate acquisition unit 7a acquires content located at a predetermined distance r from the content DB 2 based on the content Ki that was viewed immediately before the start of the search. The candidate confirmation unit 7b determines whether or not the acquired content number N is greater than a predetermined number n. When the number of contents N is larger than the predetermined number n, the center-of-gravity narrowing unit 7c divides the circumferential area of the distance r into n to narrow down to n, finds the center of gravity of each area, and is positioned at the center of gravity. Extract content.

(Hardware configuration)
FIG. 2 is a diagram illustrating a specific hardware configuration example of the content extraction apparatus. In FIG. 2, receiving means 1 receives a digital broadcast signal, and content storage means 2 stores content (such as a television program). The input means 3 includes a remote controller, a sensor for detecting the orientation of the face, line of sight, body temperature, heartbeat, and the like, and detects the user's behavior. In addition to performing system control in accordance with a built-in system program stored in the work memory 5, the CPU 4 executes the processing program in the content arrangement coordinate calculation means 6 and the processing program in the content extraction means 7 to process on the work memory 5. Create the result. The work memory 5 is a so-called working memory that temporarily stores display-related data, data used for calculations, calculation results, and the like, and stores programs, processing data, and the like used in the program processing executed by the CPU 4.

  The program may be stored in a ROM (Read Only Memory) not shown. A part of the RAM (Random Access Memory) used as the work memory 5 is backed up by a power source or is composed of a nonvolatile memory such as an EEPROM (electrically erasable programmable ROM) or a flash memory. The setting conditions are retained afterwards.

  The content display means 8 is constituted by a liquid crystal display device (LCD) or the like having a dot matrix configuration, and displays the content stored in the content storage means 2 and the calculation processing result of the CPU 4. The external storage device (recording medium) 10 is a recording medium that includes a power-backed up memory card, a DVD-ROM, or the like and records a content extraction processing program or the like. The external storage driver 9 is a device that reads the contents of the external storage device 10.

(Flowchart of content extraction process)
Hereinafter, the operation of the content extraction unit will be described with reference to a flowchart with reference to FIG.

  When the content extraction process is started, the behavior detection unit 21 determines whether or not to detect a user search start (step S100). If the search is not started (NO), the content extraction process ends (END). When the search is started (YES), the arrangement coordinate acquisition unit 7a acquires the content located at the distance r from the content Ki viewed immediately before the search from the content DB 2 (step S101). At this time, the distance r may be set in advance by the system or may be specified by the user at the time of search.

  Next, the candidate confirmation unit 7b determines whether or not the number of contents is N> n. When the acquired content number N is equal to or less than a predetermined number n (for example, 8) (NO), all the acquired content is output to the content display unit, and the process is terminated (END). On the other hand, if the acquired content number N is greater than a predetermined number n (for example, 8) (YES), the process proceeds to the next step for narrowing down candidates.

  Next, the center-of-gravity narrowing unit 7c divides the circumferential region with the distance r centered on the content Ki into a predetermined number n (for example, 8) (step S103), obtains the center-of-gravity position of each of the n-divided regions, Is extracted (step S104), and the process returns to step 102. The candidate confirmation unit 7b confirms that the acquired content number N is equal to or less than the predetermined number n (NO), outputs the extracted content to the content display unit 20, and ends the process (END). Next, a content extraction apparatus according to a second embodiment of the present invention will be described with reference to the drawings. The present embodiment is an embodiment corresponding to the solving means (2).

  In the second embodiment of the present invention, when there are more than a predetermined number n (for example, 8) of contents having the same degree of similarity, the user's viewing is performed when n representative contents are extracted from the contents. It is characterized by reflecting behavior.

  For this purpose, an evaluation value that reflects the viewing behavior of the user is set for each content, and content that is a viewing candidate is extracted based on the evaluation value. That is, the content extraction apparatus according to the second embodiment of the present invention includes a content evaluation value calculation unit 22 as shown in FIG. 10, and the content extraction unit 7 includes an evaluation value narrowing unit 7d. And

In FIG.
The content arrangement coordinate calculation unit 6 is the same as the conventional technology (FIG. 22), and includes a content DB 2, a concept vector calculation processing unit 6c, a two-dimensional coordinate calculation processing unit 6b, and a map creation processing unit 6a. However, the map creation processing unit 6a is arbitrarily provided. In the content DB 2, content arrangement coordinates 2 b and an evaluation value 2 c are recorded together with the content 2 a.

  The behavior detection unit 21 is, for example, a sensor such as a user's remote control operation, face orientation, line of sight, body temperature, heart rate, and the like. .

  The content extraction unit 7 includes an arrangement coordinate acquisition unit 7a, a candidate confirmation unit 7b, and an evaluation value narrowing unit 7d. The arrangement coordinate book acquisition unit 7a further includes a viewing unit 7a-1 and a search unit 7a-2.

  The arrangement coordinate acquisition unit 7a acquires content located at a predetermined distance r from the content DB 2 based on the content Ki that was viewed immediately before the start of the search. The candidate confirmation unit 7b determines whether the acquired content number N is greater than a predetermined number n.

  When N is larger than n, it is necessary to narrow down to n. In this embodiment, content evaluation values are used as means for extracting n representative contents based on the similarity between contents. It is characterized by that. The content evaluation value calculation unit 22 includes an evaluation value acquisition unit 22a and an evaluation value update unit 22b.

  Next, a method using content evaluation values will be described. From the content group having a predetermined similarity (distance) with a certain content Ki (for example, content that was viewed immediately before the search), n pieces of content matching the user's preference are extracted as representative content and displayed.

  Specifically, an evaluation value that reflects the user's viewing behavior is provided for each content. When a certain content is viewed, the evaluation value of the content is increased, and the evaluation value P of the content having the same similarity (distance) as the content or in the vicinity thereof is increased. The evaluation value P is decreased as the distance from the viewed content increases (FIG. 11). As a result, a three-dimensional map like a topographic map can be created (FIG. 12). As shown in FIG. 12, a three-dimensional map of evaluation value peaks P1 to P8 is generated. However, the map may be multidimensional and need not be held in the form of a map. In addition, it is estimated that a user's preference changes with progress of time. Therefore, the evaluation value returns to the original value as time passes.

(Content extraction method)
(1) From the content group (extraction candidate) at a distance r from the content Ki, the top n content items having the largest evaluation value are extracted and displayed (FIG. 13). As a result, it is possible to select content that matches the user's preference.
(2) The circumference of the predetermined distance r is divided into n with the content Ki as the center. The content having the maximum evaluation value in each area is extracted and displayed. (When there are a plurality of contents having the maximum evaluation value in the same area, the contents are located at the center of gravity.)
(3) The circumference of the predetermined distance r is divided into n with the content Ki as the center. Among the regions including the circumference, the content having the maximum evaluation value is extracted and displayed only in the regions P11, P21, P31, and P41 including the content whose evaluation value exceeds a certain threshold value (FIG. 14).
(4) The region dividing methods (2) to (3) may be the concentric dividing method in the solving means (1).
(5) Based on the evaluation value of the content, the center of gravity position (C1, C2, C3,...) Of the resulting distribution is obtained (FIG. 15), and the content located there is extracted and displayed.

  In the following, when there are more than a predetermined number n (for example, 8) of contents having the same degree of similarity, when extracting n representative contents from the contents, a certain content Ki (for example, viewed immediately before the search) A method of extracting and displaying the content having the highest evaluation value from the content group having a predetermined similarity (distance) with the content) will be described in detail.

(Configuration example)
FIG. 10 is a block diagram showing a configuration example of a content extraction apparatus according to the second embodiment of the present invention. A content DB 2 that receives a digital broadcast signal and records content (such as a TV program), a content arrangement coordinate calculation unit 6 that calculates a similarity (distance) between the contents and calculates a two-dimensional arrangement coordinate, and a predetermined number of displays In addition to the content extraction unit 7 that extracts content and the content display unit 20 that displays the extracted content, a content evaluation value calculation unit 22 that calculates an evaluation value of the content is provided.

  In the content extraction unit 7 according to the present embodiment, the behavior detection unit 21 detects which content the user is viewing and whether the search has started. The arrangement coordinate acquisition unit 7a includes a viewing unit 7a-1 and a search unit 7a-2. Moreover, the evaluation value narrowing-down part 7d is provided. The content evaluation value calculation unit 22 includes an evaluation value acquisition unit 22a and an evaluation value update unit 22b. Furthermore, the content DB 2 is newly provided with an evaluation value DB (2c) for recording the evaluation value of the content. As described above, the present embodiment is characterized in that the evaluation value 2c reflecting the viewing behavior of the user is provided. Therefore, the behavior differs depending on whether the behavior detection unit 21 of the content extraction unit 7 detects the viewing behavior of the user or the search behavior.

(Content evaluation value calculation process = flow chart at the time of viewing)
Hereinafter, the operation when the user's viewing is detected in the second embodiment of the present invention will be described with reference to the flowchart shown in FIG.

  The action detection unit 21 always detects whether the user is viewing content or performing a search (step S200). If a search is detected, the process proceeds to a content extraction process, and the content evaluation value calculation process ends. If the viewing is detected, the viewing unit 7a-1 of the arrangement coordinate obtaining unit 7a obtains the content DB 2b from the content DB 2b with the same or similar similarity to the viewing content Ki and outputs it to the evaluation value obtaining unit 22a ( Step S201).

  Next, the evaluation value acquisition unit 22a acquires the evaluation value of the acquired content from the evaluation value DB 2c, and outputs it to the evaluation value update unit 22b (step S202). The evaluation value update unit 22b updates the evaluation value of the content (step S203) and inputs it to the evaluation value DB 2c (step S204).

(Evaluation value calculation method)
If the distance between the viewed content Ki and a certain content Kj is Di, j, the evaluation value (W) of the content Kj is given by the following equation (1).

  In the equation (1), the function g (Di, j) may be a Gaussian function or may be the following equation.

In the equation (2), the constant a is an evaluation value addition coefficient, and b is a distance from Ki.
In the equation (1), constants a and b may be used as variables, and may be changed according to the number of viewing times t (repetition) of content having the same similarity. (If the number of repetitions increases, the evaluation value addition rate (corresponding to a) is reduced, or the application range of evaluation value addition (corresponding to b) is narrowed.)
By optimizing the application range of the evaluation value addition (corresponding to b), it is expected that the evaluation value reflects the user's preference more appropriately.

  In addition, it is estimated that a user's preference changes with progress of time. Therefore, the evaluation value returns to the original value as time passes. Evaluation value DB2c in content DB2 takes the form of a table as shown in Table 1, for example.

  As shown in Table 1, contents and evaluation values based on A, B, and C of the contents are tabulated.

(Content extraction process = Flowchart during search)
Next, in the second embodiment of the present invention, an operation when a user search is detected will be described with reference to a flowchart (FIG. 18). When the content extraction process is started and the behavior detection unit 21 detects a user search (step S210), the search unit 7a-2 of the arrangement coordinate acquisition unit 7a is located at a distance r from the content Ki that was viewed immediately before the search. The content to be acquired is acquired from the content DB 2 (step S211). At this time, the distance r may be set in advance by the system, or may be specified by the user at the time of search.

  Next, when the acquired content number N is equal to or less than a predetermined number n (for example, 8) (NO), the candidate confirmation unit 7b outputs all the acquired content to the content display unit 20 and ends the process. To do. On the other hand, when the acquired content number N is larger than a predetermined number n (for example, 8) (YES), the process proceeds to the next step for narrowing down candidates (step S212).

  Next, the evaluation value narrowing-down unit 7d acquires the N evaluation values of the content acquired in step S211 from the evaluation value DB 2c (step S213), and extracts the top n content having the highest evaluation value (step S214). Return to step S212. The candidate confirmation unit 7b confirms that the acquired content number N is equal to or less than the predetermined number n (NO), outputs the extracted content to the content display unit 20, and ends the process (END). The viewed content Ki itself may not be extracted as the representative content at the next search.

(Processing method when new content is recorded)
Each time a newly broadcast content is received, the similarity between content is calculated and the coordinates of the content are determined. At this time, the evaluation value of the new content Kn may be an average value of the evaluation values of neighboring content. Heretofore, evaluation values have been assigned to each content. However, the entire content map may be divided into a predetermined number of regions (for example, classification categories), and the evaluation value may be assigned to each region. At this time, the evaluation value of the new content Kn is the evaluation value of the region including the content Kn. Thereby, the evaluation value of the new content is uniquely determined.

  Next, a content extraction technique according to the third embodiment of the present invention will be described with reference to the drawings.

  In the third embodiment, in the second embodiment, when calculating the content evaluation value, the behavior is changed for each pattern (for example, time zone (morning / daytime / night), day of the week (weekday / weekend), season The evaluation value of the content is calculated separately for each location (moving / home).

  Therefore, when detecting a user's action, a user's action pattern (time, place, etc.) is detected. When the user views the content, the content evaluation value is calculated for each pattern. On the other hand, when a user searches, content is extracted and displayed using the corresponding evaluation value for each pattern. Thereby, the user can select, for example, a program that suits his / her taste, which varies depending on the time of day. The user behavior pattern includes, for example, a time zone (morning / daytime / night), day of the week (weekday / weekend), season, place (moving / home), and the like.

(Configuration example)
FIG. 19 is a block diagram showing a configuration example of a content extraction apparatus according to the third embodiment of the present invention. A content DB 2 that receives a digital broadcast signal and records content (such as a TV program), a content arrangement coordinate calculation unit 6 that calculates a similarity (distance) between the contents and calculates a two-dimensional arrangement coordinate, and a predetermined display to the user A content extraction unit 7 that extracts a number of contents, a content display unit 20 that displays the extracted content, and a content evaluation value calculation unit 22 that calculates an evaluation value of the content are provided.

  The behavior detection unit 21 in the present embodiment newly includes a behavior pattern analysis unit 21a that detects a situation (time, place, etc.) at the time of viewing and searching by the user and analyzes the behavior pattern. Further, the content DB 2 is characterized by storing an evaluation value 2d for each behavior pattern.

(Content evaluation value calculation process = flow chart at the time of viewing)
Hereinafter, in the embodiment of the present invention, an operation when a user's viewing is detected will be described with reference to a flowchart (FIG. 20).

  When the content evaluation value calculation process is started (START), the behavior detection unit 21 always detects whether the user is viewing the content or performing a search (step S300). If it is detected that the search is performed, the process proceeds to the content extraction process, and the evaluation value calculation process ends. If the viewing is detected, the behavior pattern analysis unit 21a of the behavior detection unit 21 analyzes the situation of the user at the time of viewing (current time, place, etc.), and outputs it to the viewing unit 7a-1 of the arrangement coordinate acquisition unit 7a ( Step S301).

  Next, the viewing unit 7a-1 of the arrangement coordinate obtaining unit 7a obtains the content DB2b from the content DB 2b, and obtains the evaluation value obtaining unit 22a together with the analysis result of the user situation. (Step S302).

  Next, the evaluation value acquisition unit 22a acquires the evaluation value of the acquired content from the evaluation value DB 2d of the action pattern (step S303), and the evaluation value update unit 22b updates the evaluation value of the content (step S304). ) And input to the evaluation value DB 2d for each action pattern (step S305).

(Content extraction process = Flowchart during search)
Next, an operation when a user search is detected in the third embodiment will be described with reference to a flowchart (FIG. 21).

  When the behavior detection unit 21 detects a search behavior (YES in step S310), the behavior pattern analysis unit 21a of the behavior detection unit 21 analyzes the user's situation (current time, place, etc.) at the time of the search, and an arrangement coordinate acquisition unit It outputs to the search part 7a-2 of 7a (step S311). The search unit 7a-2 of the arrangement coordinate acquisition unit 7a acquires the content located at the distance r from the content Ki viewed immediately before the search from the content DB 2 (step S312). At this time, the distance r may be set by the system in advance, or may be specified by the user at the time of search.

  Next, when the acquired content number N is equal to or less than a predetermined number n (for example, 8) (NO), the candidate confirmation unit 7b outputs all the acquired content to the content display unit 20. On the other hand, when the acquired number of contents N is larger than a predetermined number n (for example, 8) (YES), the process proceeds to the next step for narrowing down candidates (step S313).

  Next, the evaluation value narrowing-down unit 7d acquires the evaluation values of the content N acquired in step S312 from the corresponding behavior pattern (behavior pattern analyzed at the time of search) 2d in the evaluation value DB 2 (step S314), and the evaluation value The top n pieces of content having a higher value are extracted (step S315), and the process returns to step 313. The candidate confirmation unit 7b confirms that the acquired content number N is equal to or less than the predetermined number n (NO), outputs the extracted content to the content display unit 20, and ends the process (END).

  As described above, according to the content extraction device according to the present embodiment, in the present invention, the user can select content from content candidates extracted and displayed based on the similarity between content, Viewing candidate selection becomes easy.

  By extracting a certain number of viewing candidates from a content map created based on the similarity between contents according to certain rules, the similarity between recorded contents is poor or the similarity is biased Even in this case, viewing candidate selection is facilitated.

  Also, by giving each content an evaluation value that reflects the user's viewing behavior, it is possible to extract and display the content according to the user's preference, thereby increasing the user's willingness to view.

  Without creating a user profile, content adapted to the user's preferences can be extracted and displayed to the user.

  Moreover, since the original map (similarity between contents) is not changed (only the evaluation value of the contents is changed), a smaller amount of calculation is sufficient. There is an advantage that a map for each user can be created without changing the original map (similarity between similarities).

  Even if the user does not perform the evaluation work on the content, the user's preference can be accurately grasped. By changing the evaluation value back to the original value with the passage of time, it is possible to cope with a change in user's preference. This is convenient because a map for each user's action pattern can be created without changing the original map (similarity).

  The present invention can be used for a content reproduction apparatus.

It is a figure which shows one structural example of the content extraction apparatus by the 1st Embodiment of this invention. It is a figure which shows the structural example of the concrete hardware of the content extraction apparatus of FIG. It is a figure which shows the example of a display of content space. It is a figure which shows the example which divided equally the circumference 43 of the predetermined distance r into n pieces (by L). It is a figure which shows the example which equally divided the circumference L into n pieces. It is the figure divided into every cluster by clustering by the K-means method etc. based on the Euclidean distance between the semantic vectors which content has. It is a figure which shows the example divided into n concentrically centering on content Ki (1)). FIG. 6 is a diagram in which the circumference is further divided into m by L with the content Ki as the center after the concentric division. It is a figure which shows the example which displays the content (total n pieces) located in the gravity center of each area | region around the content Ki currently seen in the screen center. It is a block diagram which shows the example of 1 structure of the content extraction apparatus which concerns on the 2nd Embodiment of this invention. It is a figure which shows the distance dependence of the evaluation value of a content. It is a three-dimensional map of evaluation value peaks P1 to P8. It is a figure which shows the example which extracted and displayed the top n content with the big evaluation value value from the content group (extraction candidate) of distance r from the content Ki. It is a figure which shows the example which extracted and displayed the content with the largest evaluation value only in area | regions P11, P21, P31, and P41 including the content whose evaluation value exceeds a certain threshold value. It is the figure which calculated | required the gravity center position (C1, C2, C3, ...) of the distribution obtained based on the evaluation value of content. It is a flowchart figure which shows operation | movement of a content extraction part. It is a flowchart figure which shows operation | movement when the user's viewing is detected in the 2nd Embodiment of this invention. FIG. 10 is a flowchart illustrating an operation when a user search is detected in the second exemplary embodiment of the present invention. It is a block diagram which shows the example of 1 structure of the content extraction apparatus which concerns on the 3rd Embodiment of this invention. It is a flowchart figure which shows operation | movement when the user's viewing is detected in the 3rd Embodiment of this invention. It is a flowchart figure which shows operation | movement when a user's search is detected in the 3rd Embodiment of this invention. It is a figure which shows the structural example for demonstrating the concept of the prior art which classify | categorizes and arranges a content. It is a flowchart figure which shows the flow of the process of FIG. It is an example of spatial arrangement of contents displayed when the similarity between recorded contents is generally poor. It is a figure which shows the example in which many content groups comprised by a small number of content generate | occur | produce. It is a figure which shows the example which a very few content group which contains many contents in a specific position arises.

Explanation of symbols

  A ... Content extraction device, 2 ... Content DB, 6a ... Map creation processing unit, 6b ... Two-dimensional coordinate calculation processing unit, 6c ... Concept vector calculation processing unit, 7 ... Content extraction unit, 7a ... Arrangement coordinate acquisition unit, 7b ... Candidate confirmation part, 7c ... Center of gravity narrowing part, 20 ... Display part, 21 ... Action detection part.

Claims (25)

A content DB that stores content, a concept vector calculation processing unit that calculates a similarity vector (distance) between the content and calculates a concept vector, and a combination of all two contents using the concept vector N (n is a natural number greater than or equal to 2) dimensional coordinate calculation processing unit that calculates the distance of the content, and arrangement of the content calculated by the n-dimensional coordinate calculation processing unit based on the coordinates, the contents extraction device comprising a content extractor for extracting content to be viewing candidate, a display unit, a
Based on the content arrangement coordinates calculated by the n-dimensional coordinate calculation processing unit, the distance on the n-dimensional arrangement coordinates is within a predetermined distance r ± Δ1 (Δ1 is 0 <Δ1 ≦ r / 2). It is possible to obtain a certain number (n) of content from the content to be acquired from the content DB as viewing candidate content and based on the content arrangement coordinates calculated by the n-dimensional coordinate calculation processing unit. A content extraction unit ,
Content extraction device characterized that you view the contents extracted by the contents extracting unit on the display unit. The front Symbol display unit, contents extraction system according to claim 1, characterized in that to display the thumbnails of the content. As one of the first methods for narrowing the content extracted by the content extraction unit to n preset values,
The area having a distance r ± Δ1 on the n-dimensional arrangement coordinate is divided into n from a certain content Ki, the center of gravity of each divided area is obtained, and the content located at the center of gravity is narrowed down to n. Item 3. The content extraction device according to Item 1 or 2. As one of the first methods for narrowing the content extracted by the content extraction unit to n preset values,
From a certain content Ki, an area having a distance r ± Δ1 on the n-dimensional arrangement coordinates is clustered in advance based on the Euclidean distance between semantic vectors of the content, and is divided for each cluster. The content extraction apparatus according to claim 1 or 2, wherein a center of gravity of the region is obtained and content located at the center of gravity is narrowed down to n. As one of the first methods for narrowing the content extracted by the content extraction unit to n preset values,
The area of the distance r ± Δ1 on the n-dimensional arrangement coordinates is divided into n parts from a certain content Ki, one is extracted from each divided area, and the contents are narrowed down to n. 2. The content extraction device according to 2. As one of the second methods for narrowing the content extracted by the content extraction unit to n preset values,
A plurality of content groups classified by n types of distances are obtained based on the content arrangement coordinates based on a certain content, extracted one by one from each content group, and the content is narrowed down to n. The content extraction device according to claim 1 or 2 . The of et, each region including each of the plurality of content groups and m split obtains the center of gravity of the n × m each divided region is characterized by narrow the content to be located in the centroid in the n The content extraction device according to claim 6 . As a third method of narrowing down the content extracted by the content extraction unit to n preset values ,
Comprises a content evaluation value calculation unit which imparts an evaluation value reflecting the viewing behavior of the user, and an evaluation value refining unit that extracts n pieces of content based on the evaluation value in the content,
When the user views a certain content, the content evaluation value calculation unit increases the evaluation value of the content, and also increases the evaluation value of the content that has the same similarity (distance) as the content or a nearby content. The evaluation value decreases as the distance from the content increases,
The said evaluation value narrowing down part extracts content (n pieces) based on an evaluation value from the some content with the same similarity (distance) as a certain content, The Claim 1 or 2 characterized by the above-mentioned. Content extraction device. The evaluation value refining unit, the content extracting device according to claim 8, wherein among a distance between the certain content is a content group comparable, extracting the top n content evaluation value. Claim the evaluation value portion narrowing down is that the distance is comparable content groups with certain content divided into n, the evaluation value in said each divided region is characterized in that the extract only a total of n maximum content of 8. The content extraction device according to 8 . Range the evaluation value refining unit is the content group distances comparable with the content divided into n, only in a region including a content exceeding a certain threshold evaluation value, the evaluation value does not exceed the n the maximum content of The content extraction device according to claim 8 , wherein the content extraction device extracts the content by using the content extraction device. 9. The content according to claim 8 , wherein the evaluation value narrowing unit extracts n content based on the evaluation value from a plurality of content classified according to the distance between the content based on a certain content. Extraction device. 9. The content according to claim 8 , wherein the evaluation value narrowing unit extracts a total of n pieces of content having a higher evaluation value from each content group classified by distance between content classified based on a certain content. Extraction device. The evaluation value narrowing unit divides a content group into n parts according to the distance between contents based on a certain content, and extracts a total of n contents having the maximum evaluation value in each divided area. The content extraction device according to claim 8 . The evaluation value narrowing unit divides a content group having the same degree of similarity (distance) as a certain content into n parts, and the content having the maximum evaluation value only in a region including content whose evaluation value exceeds a certain threshold (total n maximum) The content extraction apparatus according to claim 8 , wherein the content extraction apparatus extracts the content. The evaluation value refining unit has a center of gravity position for each distribution of mass which is formed by the evaluation value of (n) calculated, to claim 8, characterized in that to extract the content (total of n) located on the center of gravity The content extraction device described. As another method for calculating a content evaluation value, the content evaluation value calculation unit calculates a content evaluation value by dividing user behavior for each pattern, and calculates an evaluation value for each pattern corresponding to the behavior pattern at the time of user search. The content extraction apparatus according to claim 8 , wherein content is extracted by using the content extraction method. The content extraction apparatus according to claim 8 , wherein the content evaluation value calculation unit is calculated so as to return to the original value as time elapses.   The content extraction apparatus according to claim 1 or 2, wherein the content extraction unit obtains a similarity based on content that was viewed at the start or immediately before the search. A content DB that stores content, a concept vector calculation processing unit that calculates a similarity vector (distance) between the content and calculates a concept vector, and a combination of all two contents using the concept vector N (n is a natural number greater than or equal to 2) dimensional coordinate calculation processing unit that calculates the distance of the content, and arrangement of the content calculated by the n-dimensional coordinate calculation processing unit A content extraction method using a content extraction device having a content extraction unit that extracts content that is a viewing candidate based on coordinates and a display unit ,
Based on the content arrangement coordinates calculated by the n-dimensional coordinate calculation processing unit, the content extraction unit has a predetermined distance r ± Δ1 (Δ1 is 0 <Δ1 ≦ r / 2) acquiring content from the content DB as viewing candidate content from content located in the range of 2);
The content extraction unit
Extracting a certain number (n) of contents based on the arrangement coordinates of the contents calculated by the n-dimensional coordinate calculation processing unit;
Displaying the content extracted by the content extraction unit on the display unit . A first method for narrowing down the content extracted by the content extraction unit to n preset values is as follows:
The method includes a step of dividing a region of distance r ± Δ1 on the n-dimensional arrangement coordinates from a certain content Ki into n, obtaining a centroid of each divided region, and narrowing down the content located at the centroid to n. The content extraction method according to claim 20. A second method for narrowing down the content extracted by the content extraction unit to n preset values is as follows:
Based on a certain content, a plurality of content groups classified by n types of distances are obtained based on the arrangement coordinates of the content, and each region including each of the plurality of content groups is divided into m, and n 21. The content extraction method according to claim 20, wherein a centroid of each of the divided areas is obtained, and n pieces of content located at the centroid are extracted. A third method of narrowing down the content extracted by the content extraction unit to n preset values is as follows:
A content evaluation value calculating step for giving an evaluation value reflecting the user's viewing behavior to the content, and an evaluation value narrowing step for extracting n pieces of content based on the evaluation value;
In the content evaluation value calculating step, when a user views a certain content, the evaluation value of the content is increased, and the evaluation value of the content having the same similarity (distance) as that of the content or a nearby content is increased, that is, the content is viewed The evaluation value decreases as the distance from the content increases,
21. The content extraction according to claim 20 , wherein in the evaluation value narrowing-down step, content (n) is extracted from a plurality of contents having the same similarity (distance) as a certain content based on the evaluation value. Method. A program for causing a computer to execute the content extraction method according to any one of claims 20 to 23 . A computer-readable recording medium for recording the program according to claim 24 .

Images