JP2014241498A - Program recommendation device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To implement program recommendation sufficiently reflected with user preference.SOLUTION: A program recommendation device (10) comprises: a topic presentation part (11) in which a topic including a predetermined keyword is generated and presented to a user; a voice analyzing part (12) which performs voice analysis on a voice of the user and extracts rhythm features of the voice; a preference degree evaluation part (15) for evaluating a preference degree of the user with respect to the keyword included in the topic on the basis of the rhythm features extracted by the voice analyzing part for the voice of the user responding to the topic; a preference DB construction part (16) for constructing a preference information database (103) in which the keyword is made correspondent to the preference degree of the user, on the basis of a result of the evaluation by the preference degree evaluation part regarding various keywords; and a program recommendation part (17) which finds and recommends a program meeting the preference of the user from program guide information while referring to the preference information database.

Description

  The present invention relates to a program recommendation device used in a digital broadcast receiver such as a digital television or a digital video recorder.

  One of the functions of a digital broadcast receiver is a program recommendation function. A typical program recommendation function is to register a keyword or genre that the user likes in the device in advance, and the device finds a program that matches the keyword or genre from program guide information such as EPG (Electronic Program Guide). And recommending it to the user (for example, automatic recording reservation). However, registering keywords, genres, etc. and updating them frequently is a cumbersome task for the user.

  Several techniques have been proposed so far to learn user viewing preferences without registering such keywords. For example, there is one that records a device operation history by a user and learns user preferences based on what device operation the user has performed for each program (see, for example, Patent Document 1). In addition, there is an apparatus that detects the orientation of the face of the user who is viewing the content, calculates the section gaze degree, and calculates the program preference degree based on the section gaze degree (see, for example, Patent Document 2).

JP 2004-194108 A JP 2011-71795 A

  The program recommendation based on the genre can reflect the user's preference only by the general classification of the genre, whereas the program recommendation based on the keyword can more accurately reflect the user's preference. According to the preference evaluation based on the operation history of the device and the section gaze degree according to the prior art, it is possible to evaluate the preference degree for the program being watched by the user, that is, the preference degree of the genre of the program. It is difficult to evaluate user preference. For this reason, there is a possibility that the user's preference cannot be sufficiently reflected by the conventional program recommendation.

  In view of the above problems, the present invention provides a program recommendation device that learns a user's viewing preference and imposes a program recommendation that fully reflects the user's preference without forcing the user to enter a keyword or the like. Objective.

  A program recommendation device according to an aspect of the present invention includes a topic presentation unit that generates a topic including a predetermined keyword and presents it to a user, and a voice analysis that analyzes a user's voice and extracts prosodic features of the voice And a preference level evaluation unit that evaluates a user's preference level for a keyword included in the topic based on prosodic features extracted by the speech analysis unit with respect to a voice that the user responds to the topic, and various On the basis of the evaluation result by the preference level evaluation unit for keywords, a preference DB construction unit that builds a preference information database in which keywords and user preference levels are associated, and a program guide with reference to the preference information database And a program recommendation unit for finding and recommending a program that matches the user's preference from the information.

  According to this, the user's preference level for the keyword included in the topic is evaluated from the prosodic features of the voice that the user responds to the topic presented by the topic presentation unit, and the keyword and the user's preference level are associated with each other. Then, a preference information database is constructed, and a program that matches the user's preference is recommended with reference to the constructed preference information database. Therefore, the user's viewing preference can be learned through interaction with the user without imposing a burden on the user such as inputting a keyword.

  The program recommendation device may include a voice recognition unit that recognizes a user's voice and extracts a keyword, and the preference evaluation unit extracts the voice freely uttered by the user by the voice analysis unit. A user's preference degree with respect to the keyword extracted by the speech recognition unit may be evaluated based on prosodic features.

  The program recommendation device may include an image analysis unit that recognizes a user's gesture by performing image analysis on an image captured of the user, and the preference evaluation unit is configured to respond to the topic recognized by the image analysis unit. In consideration of the user's gesture, the user's preference degree for the keyword included in the topic may be evaluated.

  The topic generation unit can present the generated topic to the user as voice information and / or text information.

  According to another aspect of the present invention, a program recommendation device generates a topic including a predetermined keyword and presents it to a user, and performs image analysis on an image obtained by photographing the user's gesture on the topic. Evaluation results by the image analysis unit that extracts the features of the user, the preference level evaluation unit that evaluates the user's preference level for the keyword included in the topic based on the feature of the gesture, and the preference level evaluation unit for various keywords A preference DB construction unit for constructing a preference information database in which keywords and user preference levels are associated with each other, and referring to the preference information database, find a program that matches the user's preference from program guide information. And a program recommendation section for recommending.

  According to this, the user's preference level for the keyword included in the topic is evaluated from the feature of the user's gesture for the topic presented by the topic presentation unit, and the keyword and the user's preference level are associated with each other in the preference information database. And a program that matches the user's preference is recommended with reference to the constructed preference information database. Therefore, the user's viewing preference can be learned through interaction with the user without imposing a burden on the user such as inputting a keyword.

  According to the present invention, it is possible to learn a user's viewing preference without forcing the user to enter a keyword or the like, and to make a program recommendation that sufficiently reflects the user's preference.

The block diagram of the principal part of the program recommendation apparatus which concerns on one Embodiment of this invention. Block diagram of speech analysis unit according to an example Block diagram of voice recognition unit according to an example Block diagram of the image analysis unit according to an example

  DESCRIPTION OF EMBODIMENTS Hereinafter, embodiments for carrying out the present invention will be described with reference to the drawings. In addition, this invention is not limited to the following embodiment.

  A program recommendation device according to an embodiment of the present invention is a device that learns a user's viewing preference through interaction with the user and recommends content (program) that matches the user's preference. The program recommendation device is a TV device, a video recorder, a terminal device, or the like that receives a broadcast wave such as a terrestrial digital wave or a satellite digital wave and / or a program distributed on a network such as the Internet from a broadcaster or a content distribution company To provide program recommendation functions to these devices.

  FIG. 1 is a block diagram of a main part of a program recommendation device 10 according to an embodiment of the present invention. The program recommendation device 10 according to the present embodiment includes a topic presentation unit 11, a voice analysis unit 12, a voice recognition unit 13, an image analysis unit 14, a preference level evaluation unit 15, a preference DB construction unit 16, and a program recommendation unit 17. ing.

  The topic presentation unit 11 generates various topics including a predetermined keyword and provides the user with a topic to realize interaction with the user. The topic presentation unit 11 can synthesize the generated topic by voice synthesis to generate an audio signal, and can output the topic as voice information from a speaker (not shown) and present it to the user. Further, the topic presentation unit 11 can convert the generated topic into text, generate a video signal for OSD (On Screen Display), etc., output the topic as character information from a display device (not shown), and present it to the user. .

  The topic presentation unit 11 can acquire a keyword suitable for generating a topic with reference to the keyword database 101. The keyword database 101 stores keywords suitable for searching for user viewing preferences. Keywords stored in the keyword database 101 include, for example, words representing program genres such as “drama”, “news”, “sports”, facility names such as “Tokyo Skytree”, place names, and person names. Etc.

  The program recommendation device 10 can extract keywords from an electronic program guide or the like and add them to the keyword database 101. As will be described later, the electronic program guide is stored in the program guide information database 104. Further, if the program recommendation device 10 is connected to the Internet, the latest keyword is added to the keyword database 101 by a push-type service and / or by obtaining the latest attention word from a specific web site on the Internet. Can do. Further, the program recommendation device 10 can appropriately delete keywords that are no longer used from the keyword database 101.

  The voice analysis unit 12 receives a user's voice collected by a microphone (not shown), analyzes the voice, and extracts prosodic features. Prosodic features include speech inflection, tone, stress, tone length, rhythm, speech speed, and the like. The voice analysis unit 12 can quantify such prosodic features.

  FIG. 2 is a block diagram of the speech analysis unit 12 according to an example. The voice analysis unit 12 includes, for example, a voice feature extraction unit 121 and a prosody parameter generation unit 122.

  The voice feature extraction unit 121 receives A / D converted voice data. The voice feature extraction unit 121 analyzes the input voice data and extracts a voice section, and calculates a fundamental frequency F0 for each frame, for example, with one frame as 10 ms. Moreover, the speech feature extraction unit 121 performs phoneme alignment to identify the vowel section of the first mora.

  Based on the analysis result of the speech feature unit 121, the prosody parameter generation unit 122 has 1) the slope of the fundamental frequency F0 in the vowel section of the first mora, 2) the range of the fundamental frequency F0 in the entire speech section, and 3) At least three parameters of the duration of the last phoneme in the speech section are generated. The prosody parameter generation unit 122 further includes 4) a secondary coefficient of the secondary regression coefficient in the entire speech section, 5) an average value of the entire speech section of the fundamental frequency F0 normalized by taking an average for each user, and 6) Maximum values and 7) minimum values can also be added to the prosodic features.

  Returning to FIG. 1, the voice recognition unit 13 receives a user's voice collected by a microphone (not shown), recognizes the voice and extracts a keyword.

  FIG. 3 is a block diagram of the speech recognition unit 13 according to an example. The voice recognition unit 13 includes, for example, a voice processing unit 131, a morpheme analysis unit 132, and a keyword extraction unit 133.

  The audio processing unit 131 receives A / D converted audio data. The voice processing unit 131 analyzes the input voice data and extracts a voice section, and calculates a mel frequency cepstrum coefficient (MFCC) for each frame, for example, with one frame as 10 ms. Furthermore, the voice processing unit 131 converts the voice data into text using a finite state transducer (FST).

  The morphological analysis unit 132 performs morphological analysis on the text data output from the speech processing unit 131, and decomposes the text into word groups.

  The keyword extraction unit 133 extracts specific parts of speech, particularly nouns, from the processing result of the morphological analysis unit 132. For example, the keyword extraction unit 133 may extract keywords registered in the keyword database 101 with reference to the keyword database 101.

  Returning to FIG. 1, the image analysis unit 14 receives a user's image taken by a camera (not shown), analyzes the image, and recognizes the user's gesture. The gesture recognized by the image analysis unit 14 is, for example, a movement of the user's head capable of estimating a positive (positive) or negative (negative) reaction of the user such as a nod, a shadow, and a head swing.

  FIG. 4 is a block diagram of the image analysis unit 14 according to an example. The image analysis unit 14 includes, for example, a region extraction unit 141, an optical flow extraction unit 142, a region average processing unit 143, and a gesture recognition unit 144.

  The region extraction unit 141 extracts the user's head region from the input video data. The head region can be extracted from the background image by using, for example, skin or hair color.

  The optical flow extraction unit 142 extracts the optical flow of the user's head region extracted by the region extraction unit 141. Specifically, the optical flow extraction unit 142 can extract an optical flow by a gradient method.

  The area average processing unit 143 calculates the average value of the optical flows in each quadrant for the first to fourth quadrants obtained by dividing the user's head area into four vertical and horizontal sections based on the center of gravity of the head area. The average value represents the overall flow of each quadrant.

  The gesture recognition unit 144 recognizes the user's gesture by specifying the movement of the user's head region based on the average value of the optical flow in each quadrant output from the region average processing unit 143. For example, when the first to fourth quadrants move up and down in the same manner, the gesture recognition unit 144 can recognize that the user is nodding. When the first to fourth quadrants move in the same way from side to side, the gesture recognition unit 144 can recognize that the user is shaking his / her head. When the first quadrant and the third quadrant or the second quadrant and the fourth quadrant move symmetrically around the origin, the gesture recognition unit 144 can recognize that the user is raising his / her neck.

  Returning to FIG. 1, the preference level evaluation unit 15 evaluates the user's preference level for keywords included in the topic according to the user's voice and gesture for the topic presented by the topic presentation unit 11. Specifically, the preference level evaluation unit 15 is based on the prosodic features extracted by the voice analysis unit 12 and / or the user's gesture recognized by the image analysis unit 14, and the user's preference for the keywords included in the topic Assess degree.

  When a topic is presented to the user, the user is considered to have some positive or negative response to the topic by voice. For example, when the topic “How about XX?” (XX is a predetermined keyword) is presented, if the user is interested in the keyword included in the topic, the user utters “Like” If you are not interested in it, you may utter something like “XX”. Regardless of the user's individual utterance content, in the case of a positive utterance, the slope of the fundamental frequency F0 in the vowel section of the first mora is relatively large, the range of the fundamental frequency F0 in the entire speech section is relatively wide, and the speech There is a tendency that the duration of the last phoneme in the section is relatively short. On the other hand, in the case of a negative utterance, the opposite tendency is seen. Therefore, the preference level evaluation unit 15 can evaluate whether the user's reaction is positive or negative based on the prosodic features of the user's voice.

  Specifically, the preference level evaluation unit 15 uses a support vector machine (SVM) or the like to evaluate whether the user's reaction is positive or negative. In order to use the support vector machine, prosodic features related to various user voice responses are modeled as affirmative / negative models 102.

  Also, when a topic is presented to the user, the user may show some positive or negative response to the topic by a gesture. For example, when the topic “How about XX?” (XX is a predetermined keyword) is presented, the user nods if the user is interested in the keyword included in the topic, and the head if not interested You may bend or shake your head. Therefore, the preference level evaluation unit 15 can evaluate whether the user's reaction is positive or negative based on the user's gesture.

  When the user response includes both voice and gesture, the preference level evaluation unit 15 can comprehensively evaluate them. For example, when the prosodic feature is affirmative and the user's gesture is a nod, the preference level evaluation unit 15 evaluates that the user has a strong affirmative intention for the keyword included in the topic. Can do. On the other hand, when the prosodic feature is negative and the user's gesture is swinging, the preference level evaluation unit 15 evaluates that the user has a strong negative intention to the keyword included in the topic. be able to.

  The user's preference level evaluated by the preference level evaluation unit 15 may be expressed by a plurality of levels such as two levels of affirmation / negative, four levels of strong affirmation / affirmation / denial / strong denial, or 0 May be expressed as a probability value ranging from 1 to 1.

  The preference DB construction unit 16 constructs the preference information database 103 based on the evaluation result by the preference degree evaluation unit 15. In the preference information database 103, keywords and user preference levels are associated with each other. The preference DB construction unit 16 receives the keyword included in the topic presented to the user from the topic presentation unit 11, and receives the evaluation result of the user's preference level for the keyword included in the topic from the preference level evaluation unit 15, These are registered in the preference information database 103 in association with each other.

  In addition, the preference DB construction unit 16 can also register the user's preference degree with respect to the keyword included in the voice freely uttered by the user in the preference information database 103. Specifically, the preference level evaluation unit 15 receives a keyword included in the voice freely uttered by the user from the voice recognition unit 13, and receives the evaluation result of the user's preference level for the keyword from the preference level evaluation unit 15. And associates them with each other and registers them in the preference information database 103.

  In this way, by registering user likes and dislikes for various keywords in the preference information database 103, the program recommendation device 10 can learn the viewing preference of the user.

  The program recommendation unit 17 refers to the preference information database 103 and recommends a program that matches the user's preference. For example, the program recommendation unit 17 searches the program guide information database 104 that stores program guide information such as EPG, and finds information on programs that include many keywords that are highly preferred by the user. Then, the program recommendation unit 17 notifies the user of the broadcast schedule of the found program or makes a program recommendation by automatically making a recording reservation.

  As described above, according to the present embodiment, it is possible to learn a user's preference level for various keywords through user interaction, and to recommend a program that matches the user's preference based on the learning result. As a result, the user can be freed from the troublesome work of registering interesting keywords and genres one by one.

  Note that the number of users of the program recommendation device 10 according to the present embodiment is not limited to one specific person, and a plurality of users can share one program recommendation device 10. In that case, the program recommendation device 10 can manage preference information for each user and recommend a program that matches each user's preference.

  Moreover, learning of user preference information by the interaction between the device and the user as described above is not limited to the program recommendation device, and can be applied to other home appliances that use the user preference information for services.

DESCRIPTION OF SYMBOLS 10 Program recommendation apparatus 11 Topic presentation part 12 Voice analysis part 13 Voice recognition part 14 Image analysis part 15 Preference evaluation part 16 Preference DB construction part 17 Program recommendation part 103 Preference information database

Claims (5)

A topic presenting unit that generates a topic including a predetermined keyword and presents it to the user;
A voice analysis unit that analyzes a user's voice and extracts prosodic features of the voice;
A preference level evaluation unit that evaluates a user's preference level for a keyword included in the topic, based on the prosodic features extracted by the voice analysis unit with respect to the voice that the user responds to the topic;
A preference DB construction unit that constructs a preference information database in which keywords and user preference levels are associated with each other based on evaluation results by the preference degree evaluation unit for various keywords;
A program recommendation device comprising: a program recommendation unit that refers to the preference information database and finds and recommends a program that matches a user's preference from program guide information. It has a voice recognition unit that recognizes user's voice and extracts keywords,
The preference level evaluation unit evaluates a user's preference level for a keyword extracted by the speech recognition unit based on prosodic features extracted by the speech analysis unit with respect to speech freely uttered by a user. The program recommendation device described. An image analysis unit that recognizes the user's gesture by analyzing the image taken by the user,
The preference level evaluation unit evaluates a user's preference level for a keyword included in the topic in consideration of a user's gesture for the topic recognized by the image analysis unit. The program recommendation device described in 1.   The program recommendation device according to claim 1, wherein the topic generation unit presents the generated topic to the user as voice information and / or text information. A topic presenting unit that generates a topic including a predetermined keyword and presents it to the user;
An image analysis unit that analyzes an image of a user's gesture on the topic and extracts features of the gesture;
A preference level evaluation unit that evaluates a user's preference level for keywords included in the topic based on the features of the gesture;
A preference DB construction unit that constructs a preference information database in which keywords and user preference levels are associated with each other based on evaluation results by the preference degree evaluation unit for various keywords;
A program recommendation device comprising: a program recommendation unit that refers to the preference information database and finds and recommends a program that matches a user's preference from program guide information.

Images