JP5478146B2 - Program search device and program search program - Google Patents

Description

  The present invention relates to a program search device and a program search program for searching programs such as broadcasts. In particular, the present invention relates to a program search apparatus and a program search program for searching a stored database and acquiring a program related to a designated program.

  2. Description of the Related Art In recent years, services that allow viewers to view programs desired by viewers via a communication line such as the Internet at an arbitrary time after broadcasting by accumulating broadcast programs have begun to spread. Along with this, systems and services for recommending programs to viewers based on program metadata including information such as program genres have been proposed. As a method for selecting a program to be recommended, for example, the preference is learned based on the viewing history of the viewer, and the program is recommended based on the learning result.

  Patent Document 1 acquires program information describing program contents, extracts keywords from the program information, counts the number of appearances of the keyword and the number of appearances of different keywords belonging to the category to which the keyword belongs, A technique of calculating a similarity between a search query and a searched program based on a numerical value and outputting a specific program based on the calculated similarity is disclosed.

JP 2009-043156 A

In the technique disclosed in Patent Document 1, characters obtained from EPG information (program information) based on information such as character strings enclosed in tags and character strings, character strings registered in a dictionary, parts of speech, etc. A column is extracted as a keyword, and related programs are searched based on the keyword. However, if there is a related program that does not include a keyword, a search failure occurs.
Further, in a method of searching for a program based on a history such as collaborative filtering, a program having contents related to a designated program is not always selected.

  The present invention has been made in consideration of the above-mentioned problems, and provides a program search device and a program search program capable of accurately searching for a program whose content is similar to that of a specified specific program. Objective.

  [1] In order to solve the above-described problem, a program search device according to an aspect of the present invention includes appearance frequency information indicating the appearance frequency of the expression for each program in program information including text indicating the contents of the program, and the program A program information storage unit for storing the specific representation information included in the information and the specific representation information for storing the specific representation weight value corresponding to the specific representation in association with each other, a program information acquisition unit for acquiring the program information, and a designation An appearance frequency counting unit that counts the appearance frequency of the expression in the program information acquired by the program information acquisition unit with respect to the specified program, and the appearance frequency counting unit obtained by counting the program information related to the specified program. The appearance frequency for each expression, the appearance frequency information related to a specific search target program read from the program information storage unit, and the program information Based on the specific expression information read from the memory, the degree of co-occurrence of the expression between the designated program and the search target program is calculated, and the specific expression includes the expression in the specific expression information. A similarity calculation unit that calculates a degree of association between the designated program and the search target program by weighting the degree of co-occurrence with the specific expression weight value, and the similarity A search result presentation unit that outputs the search target program selected based on the degree of association calculated by the degree calculation unit.

According to this configuration, the program information acquisition unit acquires program information related to the designated program, and the appearance frequency counting unit counts the appearance frequency of the expression in the program information. The similarity calculation unit is based on the relationship between the appearance frequency of the designated program and the appearance frequency of the search target program (co-occurrence relationship of expressions, a scale indicating the degree of co-occurrence (for example, a function based on Okapi BM25)) In addition, the degree of relevance between the two is calculated based on the unique expression weight value (corresponding to the unique expression included in the program information) read from the unique expression information stored in advance. Thereby, it is possible to calculate the relevance based on the weight of the unique expression, not just the relevance based only on the statistical feature of the appearance frequency.
Here, the expression is a word or a word chain included in the text of the program information. In order to statistically handle the appearance frequency of expressions such as compound words, it is preferable to use, for example, an n-gram (n-gram) representing a chain of n words. At this time, only n-grams excluding n-grams including stop words (particles etc.) may be used.
The above-mentioned scale representing the degree of co-occurrence is not simply determined by the appearance frequency of the expression in the designated program and the specific search target program, but also varies depending on the appearance frequency of the expression in the entire database to be searched. . That is, the value of the scale representing the degree of co-occurrence changes depending on whether it is a common expression or a program-specific expression.
Note that the search result presentation unit may select and output only one search target program, or may select and output a plurality of search target programs. Usually, the one with a high calculated relevance is selected.
The designated program and the search target program are moving image contents such as a broadcast program and a program distributed via a communication line. Note that the designated program or the search target program may be handled as a whole program, and a so-called program is a plurality of items (the item is a unit of a semantic group in the program. For example, in a news program, Each news item is an item that is referred to here.) The frequency of appearance, the calculation of the degree of association, and the target of selection are presented for each item as a unit. May be.

  [2] Further, according to one aspect of the present invention, in the program search device, the expression is n-grams, and the appearance frequency information stored in the program information storage unit is an n-gram appearance frequency for each program. And the appearance frequency information includes an appearance frequency for each of the n-grams of a predetermined plurality of chain numbers for each program, and the appearance frequency count unit is included in the program information related to the designated program. The frequency of occurrence of each of the predetermined plurality of chain numbers for each of the n-grams is counted, and the similarity calculation unit calculates the number of the predetermined plurality of chains between the designated program and the search target program. By calculating the degree of co-occurrence of n-grams, and further performing a process of weighting the degree of co-occurrence of n-grams with a specific chain number with an n-gram weight value corresponding to the chain number And calculates the serial relevance.

  According to this configuration, the degree of association between the designated program and the search target program is calculated based on the appearance frequency of each of the plurality of chained n-grams. In addition, by appropriately setting the n-gram weight value, it is possible to prevent multiple (overly) contributions of the appearance frequency of different n-grams including the same word chain and having different chain numbers from contributing to the degree of association. Can do.

  [3] In addition, according to one aspect of the present invention, in the program search device, the expression may be the specific expression information with respect to the degree of co-occurrence of the expression between the designated program and the search target program. If the specific expression is included, the expression representing the relationship between the designated program and the search target program is related based on the result of performing the process of weighting the degree of co-occurrence by the specific expression weight value. It further includes a related label determining unit for determining as a label, and the search result presenting unit outputs the search target program selected in association with the corresponding related label.

According to this configuration, an expression characterizing the relationship between the designated program and the search target program is determined as the relationship label. The process of weighting the degree of co-occurrence of expressions between the two by the specific expression weight value may be, for example, simply multiplying the value of the scale of the degree of co-occurrence by the specific expression weight value. It is reasonable to determine the expression that maximizes the result as the relation label. Since the relationship label is a label indicating the relationship between programs, if the designated program changes, the relationship label may be different even if the search result is the same program.
By outputting the determined relation label, it is possible to explain why the program selected as the search result is selected. In addition, it is possible to present a plurality of search result programs in a format classified for each determined relation label.

  [4] Further, according to one aspect of the present invention, in the program search device described above, a process of extracting a specific expression from the program information acquired by the program information acquisition unit regarding a program to be searched is performed and extracted. A specific expression specifying unit that determines the specific expression weight value corresponding to the specific expression that has been generated, creates the specific expression information, and writes the specific expression information to the program information storage unit, and the appearance frequency counting unit includes a search target The appearance frequency of the expression in the program information acquired by the program information acquisition unit is counted, and the appearance frequency information is created and written in the program information storage unit.

With this configuration, it is possible to extract appearance frequency information and unique expression information that should be held by the program information storage unit from given program information. That is, a program database necessary for program search can be constructed.
The specific expression weight value may be determined for each specific expression. Further, for example, the type of the specific expression may be determined, and the specific expression weight value may be determined according to the type. Further, the type of the unique expression may be stored as a part of the unique expression information.

  [5] Also, the program search program according to one aspect of the present invention includes a computer in the program information and the appearance frequency information indicating the appearance frequency of the expression for each program in the program information including text representing the contents of the program. Specific expression information that stores the specific expression and the specific expression weight value corresponding to the specific expression in association with each other, a program information storage unit that stores the program information acquisition unit that acquires the program information, and a designated designation An appearance frequency counting unit that counts the appearance frequency of the expression in the program information acquired by the program information acquisition unit regarding the program, and an expression frequency obtained by the appearance frequency counting unit counting the program information related to the designated program. The appearance frequency, the appearance frequency information relating to a specific search target program read from the program information storage unit, and the program information Based on the specific expression information read from the storage unit, the degree of co-occurrence of the expression between the designated program and the search target program is calculated, and the specific expression includes the expression in the specific expression information A similarity calculation unit that calculates a degree of association between the designated program and the search target program by weighting the degree of co-occurrence with the specific expression weight value, and the similarity And a search result presentation unit that outputs the search target program selected based on the degree of association calculated by the degree calculation unit.

  According to the present invention, it is possible to accurately search for related programs in consideration of the contents of programs.

It is the block diagram which showed the function structure of the program search apparatus by one Embodiment of this invention. It is the block diagram which showed the detailed functional structure in the program acquisition part (10) by the embodiment. It is the schematic which showed an example of the electronic program guide (EPG) data which the electronic program guide data acquisition part (101) by the embodiment acquires. It is the schematic which shows the example of a screen display marked up based on the web page data (homepage data) which the web page data acquisition part (102) by the embodiment acquires. It is the schematic which shows the example of a display in the screen of the broadcast program of closed caption data. It is the schematic which shows an example of the structure of the closed caption data which the closed caption data acquisition part (103) by the embodiment acquires. It is the schematic which shows the data structure of the appearance frequency information (121) which the program information storage part by the same embodiment memorize | stores. It is the schematic which shows the data structure of the specific expression information (122) which the program information storage part by the same embodiment memorize | stores. It is a flowchart which shows the procedure of the program database production | generation process by the same embodiment. It is a flowchart which shows the procedure of the program search process by the embodiment. It is the schematic which shows an example of the screen display of the presentation of the program search result by the embodiment.

Next, an embodiment of the present invention will be described with reference to the drawings.
FIG. 1 is a block diagram showing a functional configuration of the program search device according to the present embodiment. In the figure, reference numeral 1 denotes a program search device. As illustrated, the program search device 1 includes a program information acquisition unit 10, an appearance frequency counting unit 11, a program information storage unit 12, a similarity calculation unit 15, a search result presentation unit 18, and a specific expression specifying unit. 21 and a relation label determination unit 22. The program search device 1 is connected to an external storage device 91 and a server computer 92, and is configured to be able to acquire data from these external devices by communication.

  The program information acquisition unit 10 acquires program information from an external device such as an external storage device 91 or a server computer 92 (such as a web server). Here, the program information is data including text relating to program contents, such as electronic program guide data (EPG, Electronic Program Guide), web page data, closed caption data (CC), and the like. . Note that the program information acquisition unit 10 may directly acquire electronic program guide data and closed caption data from a broadcast wave. At this time, the broadcast station transmits a broadcast wave modulated using these data.

  The appearance frequency counting unit 11 counts the appearance frequency of the expression (n-gram or the like) in the program information acquired by the program information acquisition unit 10 for each of the specified designated program and the search target program to be searched.

  The program information storage unit 12 is realized using a magnetic disk device, a semiconductor memory, or the like, and stores a program database. This program database includes data of specific expression information 122 and appearance frequency information 121. The appearance frequency information 121 represents the appearance frequency of the expression for each program in the program information. The specific expression information 122 stores at least a specific expression included in the program information and a specific expression weight value corresponding to the specific expression. Details of these data will be described later.

  The similarity calculation unit 15 displays the appearance frequency for each expression obtained by the appearance frequency counting unit 11 counting the program information related to the designated program, and the appearance frequency information 121 related to a specific search target program read from the program information storage unit 12. Similarly, based on the unique expression information 122 read from the program information storage unit 12, a measure of the degree of co-occurrence of expressions between the designated program and the search target program is calculated, and the expression is represented by the unique expression information 122. In the case of the specific expression included in the specific expression, the relevance between the designated program and the search target program is determined by weighting the scale of the degree of co-occurrence with the specific expression weight value included in the specific expression information 122. calculate. Further, when using the appearance frequency of n-grams as the appearance frequency of the expression, the similarity calculation unit 15 calculates the degree of co-occurrence of n-grams having a plurality of chain numbers, and a specific chain among them. A process of weighting the degree of co-occurrence of n-grams by an n-gram weight value corresponding to the chain number is performed.

  The search result presentation unit 18 selects a search target program having a high degree of relevance based on the relevance calculated by the similarity calculation unit 15, and outputs the selected search target program (for example, displays on the screen). It is. Further, the search result presentation unit 18 may output the selected search target program in association with the relationship label determined by the relationship label determination unit 22 described later. For example, the search result presentation unit 18 classifies the search target programs for each determined relation label, and then selects and displays a search target program having a higher relevance level.

  The specific expression specifying unit 21 performs a process of extracting a specific expression from the program information using a technique such as machine learning. In addition, the specific expression specifying unit 21 uses a machine learning technique to determine the type of the specific expression when extracting the specific expression. In addition, a specific expression weight value corresponding to an individual specific expression or a specific expression weight value corresponding to a specific expression type is determined, and these are combined and written into the program information storage unit 12 as specific expression information 122.

  Based on the degree of co-occurrence of expressions between the designated program and the search target program, the relationship label determination unit 22 has a large scale of the degree of co-occurrence in the pair of programs (that is, the relationship between the programs is determined). The expression (well expressed) is determined as a relation label. At this time, the relation label determination unit 22 may perform a process of weighting the calculation result of the scale of the degree of co-occurrence with the specific expression weight value obtained from the specific expression information 122.

  FIG. 2 is a block diagram showing an internal functional configuration of the program information acquisition unit 10. As illustrated, the program information acquisition unit includes an electronic program guide data acquisition unit 101, a web page data acquisition unit 102, and a closed caption data acquisition unit 103.

  The electronic program guide data acquisition unit 101 acquires electronic program guide data from the outside. The electronic program guide data is transmitted as a broadcast wave signal, or transmitted from a web server device such as a broadcasting station via the Internet or the like. The electronic program guide data includes information on program titles, broadcast dates and times, performers, etc., in addition to text related to program contents. However, the amount and quality of the program content data included in the electronic program guide data vary depending on the broadcasting station and the program genre. In the case of a drama or documentary program, the electronic program guide data includes detailed information describing an outline and the like.

  The web page data acquisition unit 102 acquires web page data from the outside via the Internet or the like. At this time, the web page data acquisition unit 102 may specify target web page data by performing a search process using a program name, a broadcasting station name, or the like as a key as necessary. A web server device that provides data on a homepage of a broadcasting station has information regarding the contents of a program before or after broadcasting. For example, in a current program, there is little information provided at the time of broadcasting, but detailed and abundant information regarding the contents of the program may be provided from the web server device after broadcasting. Therefore, the data acquired by the web page data acquisition unit 102 serves as a role of complementing other program information.

  The closed caption data acquisition unit 103 acquires closed caption data from an external storage device or the like. In a news program or the like, since immediacy is important, electronic program guide data and web page data often do not contain detailed information. However, the closed caption data in which the utterances of the announcer and the program performers are converted into texts is useful in this apparatus because it includes detailed information regarding the contents of the news program. Note that the closed caption data acquisition unit 103 not only acquires closed caption data for each program, but also for each item in the program (here, the item is a semantic group of topics, news items, etc. in the program). Closed caption data can also be acquired.

FIG. 3 is a schematic diagram illustrating an example of electronic program guide data acquired by the electronic program guide data acquisition unit 101. As shown in the drawing, the electronic program guide data includes text data relating to program contents.
FIG. 4 is a schematic diagram illustrating a screen display example of a web page of program information marked up based on web page data acquired by the web page data acquisition unit 102. As shown in the figure, the web page data includes text data relating to program contents.
FIG. 5 is a schematic diagram showing an example of closed caption data displayed on the screen of a broadcast program. As shown in the figure, the closed caption data is text data of utterance contents in the program.

FIG. 6 is a schematic diagram showing an example of the structure of the above closed caption data. As shown in the figure, the closed caption data is held as tabular data and includes items of date, program ID (identification information), item ID, time, and closed caption. Each row of tabular data corresponds to an individual utterance having an appropriate length. The date represents the date when the program was broadcast, and the format is “year / month / day”. The program ID is data for uniquely identifying a program. The item ID is data for uniquely identifying an item (news item or the like) in the program as necessary. In addition, when it is not necessary to divide a program into a plurality of items for management, the data of this item ID is not used. The time indicates the time when the utterance in the program is started (or the time when the closed caption starts to be displayed during broadcasting), and the format is “hour: minute: second.frame number”. The closed caption is a text of the content of the utterance.
As described above, since the closed caption data includes the program ID data, the closed caption data acquisition unit 103 can easily acquire only data of a specific program. Further, since the closed caption data includes the item ID data, the closed caption data acquisition unit 103 can easily acquire only the data of a specific item in the program.
Note that the closed caption data acquisition unit 103 may acquire closed caption data in a format that does not include an item ID from the outside, and add an item ID based on the analysis result, or the closed caption data acquisition unit 103. However, you may make it acquire the closed caption data of the format containing item ID from the outside. In the former case, the closed caption data acquisition unit 103 identifies an announcer included in the video of the program using image processing technology, recognizes a silent section having a predetermined length by voice recognition, The program is divided into a plurality of items by a method of recognizing the topic based on the result of recognizing the text and the sound of the program, and a unique item ID is assigned to each item.

FIG. 7 is a schematic diagram illustrating a data configuration of the appearance frequency information 121 stored in the program information storage unit 12. The appearance frequency information 121 is data representing the appearance frequency (number of appearances) of n-grams of words for each program (or for each item in the program). Only the appearance frequency of n-grams of one type of chain number (n) may be used, or the appearance frequency of n-grams of a plurality of types of chain numbers may be used. As an example, three types of 1-gram (unigram), 2-gram (bigram), and 3-gram (trigram) are used. Moreover, you may make it use N types from 1-gram to N-gram (N is an integer greater than or equal to 2). Note that when using a plurality of types of n-grams, the number of chains does not necessarily have to be continuous, and does not necessarily include 1-gram.
As shown in the figure, the appearance frequency information 121 is configured as tabular data, and includes program ID, item ID, chain number (n), n-gram, and frequency items. The program ID and item ID are as already described. The chain number is the number of words included in one n-gram. The n-gram is character string data that describes the connection of words corresponding to the number of chains. For example, 3-grams having a chain number of 3 are represented by connecting words with “-” such as “economy-industry-saving”. The frequency is a numerical value representing the number of appearances of the n-gram in the program (or the item in the program).

FIG. 8 is a schematic diagram illustrating a data configuration of the unique expression information 122 stored in the program information storage unit. The proper expression includes proper nouns and numerical expressions. In the present embodiment, extended named expressions (Extended Named Entities) are used. Named entity information 122 is configured as a tabular data has named entity types, weights w _ene, named entity, each data item. The unique expression type is tag data representing the type of the specific expression. For example, “PERSON” (person name), “TITLE” (title name), “ORGANIZATION” (organization name), “LIVING_THING” (organism name), “GPE”. ”(Administrative unit name),“ FACILITY ”(facility name), and the like. The weight “ _wene” is a numerical value of the weight according to the specific expression. A method for determining this weight _window and a method for using this weight _window for the program search process will be described later. The data held in the data item of the specific expression is character string data representing the specific expression itself.

  Next, a processing procedure by the program search device 1 will be described. The main processes performed by the apparatus are a program database generation process and a program search process.

FIG. 9 is a flowchart showing the procedure of the program database generation process. The procedure for generating the program database will be described below with reference to this flowchart.
First, in step S101, the program information acquisition unit 10 acquires program information from the external storage device 91, the server computer 92, or the like. At this time, the program information acquisition unit 10 appropriately selects and acquires the electronic program guide data, web page data, and closed caption data described above. Further, the program information acquisition unit 10 may acquire a plurality of types of program information (for example, electronic program guide data and web page data) for a certain program.

Next, in step S102, the appearance frequency counting unit 11 counts the appearance frequency of n-grams included in the program information acquired above.
Specifically, the appearance frequency counting unit 11 first reads the program information acquired by the program information acquisition unit 10, performs morphological analysis processing of the text of the program information using existing technology, and converts these texts into a morpheme string. Decompose. Then, n-grams having a predetermined predetermined number of chains are extracted from the morpheme string. For example, when n = 1, 2, 3 is used as the number of chains, from the part “... Ministry of Economy, Trade and Industry ...” in the text, each of “Economy”, “Industry”, 1-gram, 2-gram of “economy-industry” and “industry-province” and 3-gram of “economy-industry-province” are extracted. The appearance frequency counting unit 11 is for each program (however, it may be for each item in the program. The processing described below as “for each program” can be similarly set for each item). The frequency of appearance of n-grams is counted. The appearance frequency counting unit 11 writes the result in the appearance frequency information 121 of the program information storage unit 12. Thereby, for a certain n-gram T _n , the number of programs including that T _n (“number of items” in the case of processing for each item in the program. In the following, it is collectively referred to as “number of programs”. and similarly) m If the process is not the case with the program to be a program of each are mixed, the number tf of T _n contained in the program is obtained.
In this process, the appearance frequency counting unit 11 extracts n-grams that do not include a predetermined stop word (particle, etc.). As a result, appearance frequency information that better represents the characteristics of the program content can be obtained.

  In step S103, the specific expression specifying unit 21 extracts the specific expression included in the program information acquired above. The proper expression means a proper noun such as a person name, a place name, and an organization name and a numerical expression, and is an expression necessary for understanding Teguisudo. At this time, the specific expression specifying unit 21 adds tag data representing the type of specific expression to the extracted specific expression. As a result, the unique expression tag can be assigned to the n-gram extracted by the appearance frequency counting unit.

There are various types of proper expressions. Workshop on Information Extraction and Information Retrieval Technology IREX (Information Retrieval and Extraction Exercise) consists of organization name (ORGANIZATION), person name (PERSON), place name (LOCATION), date expression (DATE), time expression (TIME), monetary expression Eight types were defined: (MONEY), ratio expression (PERCENT), and artifact (ARTIFACT).
In addition, a document (Akira Sekine, “Sekine Extended Specific Expression Hierarchy” Definition Version 7.0.0, March 8, 2007, URL: http://nlp.cs.nyu.edu/ene/version7_0_0.html) Defines 200 specific expression tags that are conceptually hierarchized as extended specific expressions. The types of proper expressions defined in the uppermost hierarchy are names, personal names, god names, organization names, place names, facility names, product names, event names, natural object names, disease names, and color names.
The definition of the type of the specific expression may be appropriately selected and used. However, as an example, the PERSON (person name), TITLE (from the 200 types defined in the above extended specific expression may be used. Title), ORGANIZATION (organization name), LIVING_THING (biological name), GPE (Geological Political Entity, political place name (country name, prefecture name, state name, county name, city name, etc.)), FACILITY (facility name) ) Is used.

In the same step, the specific expression specifying unit 21 extracts specific expressions as follows:
Extraction methods using existing machine learning technologies such as SVM (Support Vector Machine) and CRF (Conditional Random Fields), and information given in advance such as dictionaries and rules There are methods of extracting or combining them.
An outline of a method for extracting a specific expression using machine learning technology is as follows. That is, first, morphological analysis of program information text is performed. And the feature of each morpheme obtained as a result is extracted. Here, the features are morpheme reading (obtained from dictionary data), concept (obtained from concept dictionary data), part of speech, etc. Create a dimension feature vector. As an example, a 57-dimensional feature vector was used. As a result, the problem of extracting a specific expression from text is reduced to the problem of identifying whether or not a feature vector corresponding to a certain morpheme has a specific label (whether it corresponds to a kind of extended specific expression). The Therefore, a person labels the learning data. Learning is performed by SVM or CRF using this as teacher data. By performing this learning, it becomes possible to identify whether or not a feature vector corresponding to a given morpheme has a specific label. In other words, by using the result of machine learning, it is possible to label a specific expression to a morpheme included in an arbitrary input text. That is, it is possible to extract a specific expression from the acquired program information and specify the type of the specific expression extracted at the same time.

In step S <b> 104, the specific expression specifying unit 21 determines a weight _window for the specific expression extracted above.
Specifically, the specific expression specifying unit 21 assigns a predetermined weight to each type (tag) of the specific expression. At this time, for example, when importance is placed on the name of the person, the value of the weight for the tag “PERSON” is set to be relatively large. Further, the weight value may be determined for each unique expression instead of for each kind of specific expression.
As a method for determining the weight value, the similarity between programs is evaluated manually in advance, and the result (an index indicating how similar the program A and the program B are) is used as teaching data for learning. And This teacher data is given a tag of specific expression by hand or machine learning. Then, an initial value of the weight of each unique expression is set as appropriate, and the similarity between programs is calculated by processing (to be described later) of the similarity calculation unit 15 using the weight. The calculation result is compared with the teacher data and evaluated by an evaluation function. Then, the weight is changed in the direction in which the value of the evaluation function is improved by using the steepest gradient method or the like. The optimum weight of each proper expression obtained as a result is adopted.

  The specific expression specifying unit 21 writes the results obtained in steps S103 and S104 into the specific expression information 122 of the program information storage unit 12. The above is the program database generation process.

FIG. 10 is a flowchart showing the procedure of the program search process. The program search procedure will be described below with reference to this flowchart.
First, in step S201, program designation means (not shown) designates a program Q as a search base. When the program search device 1 according to the present embodiment also has a function (television receiver function) for receiving a broadcast signal and displaying the video on the screen, this program designating means is, for example, a program currently being displayed. Is designated as the program Q. Further, the program designating unit may designate the program Q as a search base based on a signal from the outside of the program search device 1 or an operation by the operator.

In step S202, the program information acquisition unit 10 adds program information about the program Q specified above to the external storage device 91, the server computer 92, the digital data signal added to the currently received program Q, and the like. Get from. The specific program information acquisition operation by the program information acquisition unit 10 is as described in the description of step S101 in the previous figure.
Next, in step S203, the appearance frequency counting unit 11 extracts n-grams included in the data based on the program information about the program Q acquired above, and counts the appearance frequency of those n-grams. The specific operation by the appearance frequency counting unit 11 is as described in the description of step S102 in the previous figure.

  Next, in step S <b> 204, the similarity calculation unit 15 calculates the degree of association between the program Q specified above and the program previously stored in the program database of the program information storage unit 12. At this time, the similarity calculation unit 15 calculates the degree of association between the program Q and another program D by calculation represented by the following equation (1).

Here, T _n is an n-gram included in the program Q. N is the chain number of the n-gram.
Further, the _window is a weight corresponding to the extended specific expression as described above. By calculating the relevance using this weight “ _wene” , it is possible to calculate the relevance based not only on the statistical characteristics of the appearance frequency but also on the meaning of the word.
In addition, w _ng is a weight corresponding to the chain number of n-grams.
w _ng = 1 / n
Or
w _ng = 1 / (2 ⁿ )
As described above, the weight w _ng decreases as n increases. By using this weight w _ng , it is possible to prevent an excessively high degree of association from being calculated for the compound word included in the program information. For example, when the word “Ministry of Economy, Trade and Industry” is included in the program information of both program Q and program D, not only 3-gram “economy-industry-ministry” but also 2-gram “economy— “Industry”, “Industry-Ministry”, 1-gram “Economy”, “Industry”, “Ministry”, etc. co-occur between both program information. Even in such a case, the element of the weight w _ng acts in the equation (1) to prevent these n-grams from acting excessively on the relevance.

S _BM25 (T _n , D) in Expression (1) is used in a search engine or the like, and uses a function based on Okapi BM25 for ranking the relevance between words and documents. This S _BM25 (T _n , D) calculates a score in a program D of a certain n-gram T _n and is expressed by the following equation (2).

Here, M is the number of programs of all programs stored in the program information database. M is the number of programs including n-gram T _n among them. Further, tf is the frequency of the n-gram T _n in the program information of the program D, and the data is read from the appearance frequency information 121 of the program information storage unit 12. Qtf is the frequency of n-gram T _n in the program information of program Q, and the data is obtained in step S203 above.
K is expressed by the following equation (3).

Here, dl is the explanatory sentence length of the program information of the program D (that is, the number of words included in the text). Moreover, avdl is the average of the description sentence length (number of words) of the program information of all programs stored in the program information database.
Further, b, k ₁ , and k ₃ in the expressions (2) and (3) are parameters that are set as appropriate. An example of these parameters is:
b = 0.75
k ₁ = 1.2,
k ₃ = 7
The value is used.

  The similarity calculation unit 15 calculates the degree of association S (D, Q) with the program Q for all the programs D stored in the program database by the calculation method described above. Based on the score of the result, the similarity calculation unit 15 outputs an upper program having a high degree of relevance.

  In step S205, the relationship label determination unit 22 determines a relationship label indicating the relationship between the search source program Q and the program obtained as a search result. Specifically, the relation label determination unit 22 obtains a label L (D, Q) indicating a relation between the program Q as a search source and the program D stored in the program database by the following equation (4). .

That is, when the program Q and the program D are given, the relationship label determination unit 22 uses the function S _BM25 (T _n , D) expressed by the equation (2) to obtain S _BM25 (T _n , D) × w. _ene (T _n ) is calculated, and an n-gram T _n that maximizes this value is determined as a label representing the relationship between the program Q and the program D. As is clear from such a determination method, a different label may be determined for a certain program D held in the program database depending on the designated program Q.
At this time, if the label determined above corresponds to the specific expression, the related label determination unit 22 uses the information read from the specific expression information 122 of the program information storage unit 12 as the label. The corresponding unique expression type may be determined together.

  The program search device 1 may be configured not to have the related label determination unit 22. In the case of such a configuration, only the process of step S205 is omitted from the program search process described here. When the relationship label is not required, the relationship label determination unit 22 may be omitted as described above. In this case, the program search device 1 obtains the search result based only on the calculation result by the similarity calculation unit 15. Present to the operator.

  Next, in step S206, the search result presentation unit 18 presents the search result to the operator by displaying the search result on the computer screen based on the result of the series of processes described above. Specifically, the search result presentation unit 18 outputs up to the top P programs (P is a positive integer) based on the relevance score calculated above.

  FIG. 11 is a schematic diagram illustrating an example of a screen display of search result presentation by the search result presentation unit 18. In the example shown in the figure, the search result presentation unit 18 displays the designated program Q in the vicinity of the center of the screen together with the title of the program Q (“ten people”, etc.). Also, other programs having a high degree of association with the program Q are displayed around the place where the program Q is displayed. In the figure, other programs having a high degree of association with the program Q are classified for each label and displayed together with the unique expression types corresponding to the labels. If the obtained label is not a unique expression, “Topic:” is displayed. That is, the label “Kenshin” corresponds to the specific expression type “Person”, the label “Nobunaga” corresponds to the specific expression type “Person”, the label “Kanetsu” corresponds to the specific expression type “Person”, and the label “ “Gifu Castle” corresponds to the specific expression type “Facility”. In the example, the label “Gun” is not assigned a unique expression tag, and therefore corresponds to “Topic”. Of the corresponding programs corresponding to the respective labels, the programs having the highest degree of relevance with the program Q are displayed in order from the top (display thumbnail images corresponding to the respective programs in the search results). You may).

That is, at this time, the search result presentation unit 18 uses the program D, the score S (D, Q) of the degree of association between the program Q and the program D calculated by the similarity calculation unit 15, and the relationship label determination unit 22. Based on the determined relationship label L (D, Q) indicating the relationship between the program Q and the program D, the program (D) in the program database is classified for each label, and a predetermined upper number belonging to each of these labels A predetermined number of labels having the highest total score of the program (D) are determined. Then, the upper program (D) belonging to each determined label is displayed.
At this time, the search result presentation unit 18 displays the thumbnail images of the related programs described above around the video content of the program Q displayed at the center of the screen, and the operator displays those thumbnails. When an operation for clicking one of the images is performed, the selected related program is changed to display and reproduction at the center of the screen as the designated program Q.

In summary, the search result presentation unit 18 displays the search result on the screen by the method (a) below.
(A) The search result presentation unit 18 displays the designated program Q, and also includes a label having a high total relevance of the upper predetermined number of programs, a unique expression type corresponding to the label, and the label Of programs, programs with the highest degree of relevance are displayed.
In addition to this method, the search result presentation unit 18 may display the search result on the screen by any one of the methods (b) to (f) below.
(B) The search result presentation unit 18 displays the same as in (a) above, but does not display the unique expression type.
(C) In addition to the display of either (a) or (b) above, the search result presentation unit 18 does not include the total level of relevance of the programs included in the label, When there is a program that has a high degree of association with the program Q alone, such a single program or a plurality of programs are displayed together. In addition, since these programs do not have a common label, words such as “others” may be displayed together instead of the label.
(D) The search result presentation unit 18 simply selects and displays a predetermined number of programs having higher relevance S (D, Q) from all programs in the program database without considering the labels. In this case, the program search device 1 may not include the related label determination unit 22.
(E) The search result presentation unit 18 does not classify the programs into labels and arranges them, but selects a predetermined number of programs having a higher relevance S (D, Q) from all programs in the program database and scores them. Are displayed in order, and a character string with a label corresponding to each program is displayed.
(F) In addition to the display of (e) above, the search result presentation unit 18 also displays a specific expression type corresponding to the label, if any.
In addition, the search result presentation unit 18 may perform display by further variations of the above (a) to (f).

In addition, you may make it implement | achieve a part or all function of the program search apparatus in embodiment mentioned above with a computer. In that case, the program for realizing the control function may be recorded on a computer-readable recording medium, and the program recorded on the recording medium may be read by a computer system and executed. Here, the “computer system” includes an OS and hardware such as peripheral devices. The “computer-readable recording medium” refers to a storage device such as a flexible medium, a magneto-optical disk, a portable medium such as a ROM and a CD-ROM, and a hard disk incorporated in a computer system. Further, the “computer-readable recording medium” dynamically holds a program for a short time, like a communication line when transmitting a program via a network such as the Internet or a communication line such as a telephone line. It is also possible to include those that hold a program for a certain time, such as a volatile memory inside a computer system serving as a server or client in that case. The program may be a program for realizing a part of the functions described above, and may be a program capable of realizing the functions described above in combination with a program already recorded in a computer system.
Moreover, you may implement | achieve by incorporating the function of the program search device in embodiment mentioned above in a broadcast receiver.

  The embodiment of the present invention has been described in detail with reference to the drawings. However, the specific configuration is not limited to this embodiment, and includes designs and the like that do not depart from the gist of the present invention.

<Results of evaluation experiment>
In order to examine how accurately the program related to the program being viewed by the viewer can be searched, an evaluation experiment of related program search using the prototype system of the above embodiment was performed. In the experiment, 15 programs with no overlapping types were used as search sources, and the top 10 programs in the search results were evaluated. There are 4 levels of relevance between programs (4: related, 3: slightly related, 2: not very related, 1: not related), and 3 levels of relationship labels (3: appropriate, 2: (Substantially appropriate, 1: inappropriate). As a baseline, Cosine similarity with morpheme tf-idf values as features was used. A morpheme having the highest tf-idf value was assigned to the label.

  As a result of this experiment, the prototype system of the above embodiment exceeded both the relevance and the relation label compared to the baseline. Specifically, regarding relevance, the baseline evaluation result was 3.26, while that of the prototype system was 3.35. Regarding the relation label, the baseline evaluation result was 2.25, while that of the prototype system was 2.49. This can be analyzed because the prototyping system technique evaluates ENE (extended specific expressions) and compound words, and programs having meaningful vocabulary that the user judges to be “related” are gathered at the top. On the other hand, although it is a small number, the weight of the ENE that is not directly related to the content, such as the country name exemplified in the program information, may have contributed to the score. This can be improved by performing selective ENE weighting (setting of specific expression weight values) in consideration of the context of the search source content. In relation label evaluation, it can be analyzed that a meaningful word such as a compound word or ENE has been determined as a relation label, thereby improving the user's evaluation. That is, the effectiveness of the embodiment of the present invention was confirmed by experiments.

  INDUSTRIAL APPLICABILITY The present invention can be used for recommending other related programs to viewers who view broadcast programs and video programs distributed via communication lines. In addition, it is industrially useful to apply such a function to a service that provides, for example, a large number of abundant video programs based on the viewer's request.

DESCRIPTION OF SYMBOLS 1 Program search device 10 Program information acquisition part 11 Appearance frequency counting part 12 Program information storage part 15 Similarity degree calculation part 18 Search result presentation part 21 Specific expression specific | specification part 22 Relation label determination part

Claims (4)

Appearance frequency information indicating the appearance frequency of the expression for each program in the program information including text representing the contents of the program, the specific expression included in the program information, and the specific expression weight value corresponding to the specific expression are stored in association with each other. A program information storage section for storing
A program information acquisition unit for acquiring the program information;
An appearance frequency counting unit that counts the appearance frequency of the expression in the program information acquired by the program information acquisition unit with respect to the specified program;
The appearance frequency for each expression obtained by the appearance frequency counting unit counting the program information related to the designated program, the appearance frequency information related to a specific search target program read from the program information storage unit, and the program Based on the specific expression information read from the information storage unit, the degree of co-occurrence of the expression between the designated program and the search target program is calculated, and the expression is included in the specific expression information. If it is an expression, a similarity calculation unit that calculates a degree of association between the designated program and the search target program by weighting the degree of co-occurrence with the specific expression weight value;
A search result presentation unit that outputs the search target program selected based on the relevance calculated by the similarity calculation unit;
A program search device characterized by comprising :
The expression is an n-gram of a word ;
The appearance frequency information stored in the program information storage unit represents an appearance frequency of an n-gram of a word for each program, and the appearance frequency information is a predetermined plurality of chain numbers of words for each program . including the frequency of appearance every n grams,
The appearance frequency counting unit counts the appearance frequency for each n-gram of the word of the predetermined plurality of chain numbers in the program information related to the designated program,
The similarity calculation unit calculates a degree of co-occurrence of n-grams of the predetermined plurality of chain words between the designated program and the search target program, and a word having a specific chain number The degree of co-occurrence of n-grams is calculated by further performing a process of weighting the degree of co-occurrence with n-gram weight values such that the weight decreases as the number of chains increases ,
A program search device characterized by that. In contrast to the degree of co-occurrence of expressions between the designated program and the search target program, if the expression is a specific expression included in the specific expression information, the degree of co-occurrence is set as the specific expression weight. A relation label determination unit for determining, as a relation label, the expression representing a relation between the designated program and the search target program based on a result of performing a weighting process by a value;
The search result presentation unit outputs the search target program selected in association with the corresponding relation label.
The program search device according to claim 1 , wherein: A process for extracting a specific expression from the program information acquired by the program information acquisition unit for a program to be searched is performed, and the specific expression weight value corresponding to the extracted specific expression is determined, and the specific expression is determined. A unique expression specifying unit that creates expression information and writes it in the program information storage unit;
The appearance frequency counting unit counts the appearance frequency of the expression in the program information acquired by the program information acquisition unit for the program to be searched, and creates the appearance frequency information in the program information storage unit. Write,
Program search apparatus according to claim 1 or 2, characterized in that. Computer
Appearance frequency information indicating the appearance frequency of the expression for each program in the program information including text representing the contents of the program, the specific expression included in the program information, and the specific expression weight value corresponding to the specific expression are stored in association with each other. A program information storage section for storing
A program information acquisition unit for acquiring the program information;
An appearance frequency counting unit that counts the appearance frequency of the expression in the program information acquired by the program information acquisition unit with respect to the specified program;
The appearance frequency for each expression obtained by the appearance frequency counting unit counting the program information related to the designated program, the appearance frequency information related to a specific search target program read from the program information storage unit, and the program Based on the specific expression information read from the information storage unit, the degree of co-occurrence of the expression between the designated program and the search target program is calculated, and the expression is included in the specific expression information. If it is an expression, a similarity calculation unit that calculates a degree of association between the designated program and the search target program by weighting the degree of co-occurrence with the specific expression weight value;
A search result presentation unit that outputs the search target program selected based on the relevance calculated by the similarity calculation unit;
A program search program that functions as a program search device comprising :
The expression is an n-gram of a word;
The appearance frequency information stored in the program information storage unit represents an appearance frequency of an n-gram of a word for each program, and the appearance frequency information is a predetermined plurality of chain numbers of words for each program. including the frequency of appearance every n grams,
The appearance frequency counting unit counts the appearance frequency for each n-gram of the word of the predetermined plurality of chain numbers in the program information related to the designated program,
The similarity calculation unit calculates a degree of co-occurrence of n-grams of the predetermined plurality of chain words between the designated program and the search target program, and a word having a specific chain number The degree of co-occurrence of n-grams is calculated by further performing a process of weighting the degree of co-occurrence with n-gram weight values such that the weight decreases as the number of chains increases,
A program search program characterized by that.

Images