JP7606866B2 - Genre-specific text collection device and its program - Google Patents

Description

The present invention relates to a genre-specific text collection device and a program for collecting text data by genre.

Building language models for speech recognition and analyzing natural language processing requires a large amount of natural language text (hereafter referred to as a text corpus). One method for obtaining this text corpus is to collect text data posted on web pages. This method is often used to obtain a text corpus, as a large amount of information, including text data, is available on the Internet, and most of it can be accessed freely.

For example, Patent Document 1 discloses a method for collecting text data related to the speech recognition target from the Internet by selecting a word set from a text corpus of speech recognition targets collected in advance and using the word set as a search query for a search engine.

As another method, Patent Document 2 discloses a method for collecting text data for a specific purpose by extracting word strings that match a pre-prepared word string template from a text corpus containing multiple natural language sentences and converting the extracted word strings into a word string in a format suited to the purpose using conversion rules.

Patent document 3 also discloses a method for collecting subtitles of broadcast programs as learning data for a language model.

JP 2012-83543 A JP 2012-78647 A JP 2019-8315 A

The method disclosed in Patent Document 1 requires manual creation of a document format for character strings for selecting word sets. Therefore, this method requires the effort of creating a separate document format for each genre in order to collect text data by genre. In addition, this method has the problem that the collected text data depends on the document format and the accuracy of the search engine.

The technique disclosed in Patent Document 2 also requires the manual creation of word example templates and transformation rules, which is time-consuming, and has the problem that the collected text data depends on the characteristics of the word example templates and transformation rules.

The method disclosed in Patent Document 3 can collect a large amount of text data from the subtitles of broadcast programs. However, broadcast programs come in a variety of genres, including news, information, drama, and animation. Therefore, this method requires manual selection of broadcast programs by genre, making it difficult to collect a large amount of text data by genre.

The present invention was made in consideration of these problems, and aims to provide a genre-specific text collection device and a program therefor that can accurately collect large amounts of genre-specific text data without relying on document formats, templates, etc. to identify genres.

To solve the above problem, the genre-specific text collection device of the present invention is a genre-specific text collection device that collects genre-specific text from subtitle text multiplexed into digital broadcasting, and is configured to include a broadcast receiving means, a subtitle information extraction means, an EPG information extraction means, a program information identification means, and a text extraction means.

In such a configuration, the genre-specific text collection device receives digital broadcasts by the broadcast receiving means, and demodulates them into TS (Transport Stream) signals.
The genre-specific text collection device then uses the subtitle information extraction means to extract subtitle information including subtitle text and time information for presenting the subtitle text from the TS signal demodulated by the broadcast receiving means. Note that the subtitle information is multiplexed into the TS signal, and its data format is standardized by ARIB (Association of Radio Industries and Businesses). Therefore, the subtitle information extraction means can extract the multiplexed subtitle information by analyzing the data format of the TS signal.

The genre-specific text collection device also uses an EPG information extraction means to extract EPG information of broadcast programs from the TS signal. EPG information is information for generating an electronic program guide, and includes broadcast program time information, genre, etc. Like subtitle information, this EPG information is multiplexed into the TS signal, and its data format is standardized by ARIB. Therefore, the EPG information extraction means can extract the multiplexed EPG information by analyzing the data format of the TS signal.

The genre-specific text collection device then uses the program information identification means to identify the time information and genre of the broadcast program from the EPG information. Note that the genre is information set by each broadcasting station, but since it is information that identifies the broadcast content, there is little difference between broadcasting stations. Therefore, the genre extracted from the EPG information is highly accurate information for the extracted text.
A genre is composed of higher-level categories classified by higher-level items and lower-level categories subdivided into the higher-level categories. Therefore, the program information identifying means identifies only the higher-level categories as genres. Also, one or more genres are set for each broadcast program in the EPG information. Therefore, when multiple genres are set in the EPG information, the program information identifying means identifies the most frequently set higher-level category for the broadcast program as the genre.

Then, the genre-specific text collection device uses a text extraction means to extract subtitle text for a time period specified by the time information of the broadcast program from the subtitle information, and associates it with the genre of the broadcast program to generate genre-specific text.

This allows the genre-specific text collection device to extract subtitle text from the broadcast signal by genre based on the EPG information.
The genre-specific text collection device can be operated by a program that causes a computer to function as each of the above-mentioned means.

The present invention provides the following excellent effects.
According to the present invention, it is possible to collect a large amount of subtitle text by genre based on the genre and time information of broadcast programs set in EPG information.
As a result, the present invention can obtain a highly accurate text corpus classified into genres required for speech recognition, natural language processing, and the like, without requiring manual effort.

1 is a block diagram showing a configuration of a genre-specific text collection device according to an embodiment of the present invention. 1 is a data structure diagram showing a schematic structure of a transport stream (TS) signal. 11 is a diagram showing an example of subtitle information extracted from a TS signal by a subtitle information extraction means. FIG. FIG. 4 is a diagram showing an example of EPG information extracted from a TS signal by an EPG information extraction means. FIG. 2 is an explanatory diagram for explaining genre classification. 11 is an explanatory diagram for explaining an example of extracting subtitle text by genre from subtitle information in a subtitle text extraction means. FIG. FIG. 13 is an explanatory diagram for explaining the processing contents of regular expression filter processing in the shaping means; 3 is a flowchart showing the overall operation of the genre-specific text collection device according to the embodiment of the present invention. 9 is a flowchart showing a detailed operation of genre specification in FIG. 8;

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
<Configuration of genre-specific text collection device>
The configuration of a genre-specific text collection device 1 according to an embodiment of the present invention will be described with reference to FIG.

The genre-specific text collection device 1 collects genre-specific text from subtitle text multiplexed in digital broadcasting.
As shown in FIG. 1, the genre-specific text collection device 1 includes a broadcast receiving means 10, a broadcast information extracting means 11, a storage means 12, a program information identifying means 13, and a text extracting means 14.

The broadcast receiving means 10 receives and demodulates digital broadcast waves. The broadcast waves may be terrestrial digital broadcast, satellite broadcast, cable broadcast, or the like, and may be wireless or wired.
The broadcast receiving means 10 is a television tuner that receives digital broadcast waves and decodes them to convert them into MPEG-2 transport stream signals (hereinafter referred to as TS signals).

2, the TS signal is multiplexed with video and audio information 200, subtitle information 201, EPG information 202, data broadcast information 203, etc. Note that the TS signal is a signal standardized by ARIB (Association of Radio Industries and Businesses), and therefore a detailed description thereof will be omitted here.
The broadcast receiving means 10 demodulates the broadcast signal of the specified channel from the broadcast wave. The number of channels to be demodulated is not limited to one, and the broadcast receiving means 10 may be configured as a plurality of tuners.

The genre-specific text collection device 1 uses only subtitle information and EPG information in the broadcast information extraction means 11 described below. Therefore, the broadcast reception means 10 may be a television tuner that is not limited to receiving full-segment broadcasts, handles a small amount of information, does not require copy control by a CAS (Conditional Access System), and is inexpensively available and supports one-segment broadcasts.

The broadcast receiving means 10 outputs the demodulated TS signal to the broadcast information extraction means 11. When receiving multiple channels, the broadcast receiving means 10 outputs a TS signal for each channel to the broadcast information extraction means 11.

The broadcast information extraction means 11 extracts subtitle information and EPG information for a specified time interval from the TS signal demodulated by the broadcast reception means 10 .
The specified time interval is the time interval during which the user wants to collect text, and is specified externally by the collection start time (e.g., specified by year, month, day, hour, minute, and second) and the collection end time (e.g., specified by year, month, day, hour, minute, and second). Alternatively, this time interval may be the interval from when a start instruction is given by an external switch or the like to when an end instruction is given. Alternatively, this time interval may be specified by a start instruction and a time length.

The length of this time period can be freely set to one day, one week, one month, one year, etc. The broadcast information extraction means 11 includes a timing means (timer) (not shown) for timing the time period.
As shown in FIG. 1, the broadcast information extraction means 11 includes a subtitle information extraction means 110 and an EPG information extraction means 111 .

The subtitle information extraction means 110 extracts subtitle information including subtitle text and time information for presenting the subtitle text from the TS signal.
The subtitle information extraction means 110 analyzes the TS signal and extracts subtitle information for a specified time period. An example of the subtitle information extracted by the subtitle information extraction means 110 is shown in FIG.

As shown in FIG. 3, the subtitle information extracted by the subtitle information extracting unit 110 includes a date 300 , a start time 301 , and subtitle text 302 .
Date 300 is the date (year/month/day) that the corresponding subtitle text 302 will be presented on the television screen.
Start time 301 is the time (hours:minutes:seconds) when corresponding subtitle text 302 is to be presented on the television screen.
The subtitle text 302 is a character string of subtitles to be presented on the television screen at the corresponding date 300 and start time 301 .
For example, the example in FIG. 3 indicates that "Now in a cutting-edge megalopolis." will be used as a subtitle at the start time (06:00:12) on the date (2020/07/08).

The subtitle information extraction means 110 writes and stores the extracted subtitle information in the storage means 12. In the case of a TS signal corresponding to multiple channels, the subtitle information extraction means 110 extracts subtitle information from each TS signal and stores it in the storage means 12. When extracting subtitle information for multiple channels, the subtitle information extraction means 110 may store the information in separate memory areas for each channel in the storage means 12, or may store the information without distinguishing between channels.

The EPG information extraction means 111 extracts EPG (Electronic Program Guide) information of a broadcast program from a TS signal.
The EPG information extraction means 111 analyzes the TS signal and extracts the EPG information for a specified time period. An example of the EPG information extracted by the EPG information extraction means 111 is shown in FIG.

As shown in FIG. 4, the EPG information extracted by the EPG information extracting means 111 includes a date 400 , a start time 401 , a duration 402 , a genre identifier 403 , and a title 404 .
The date 400 is the date (year/month/day) on which the broadcast program of the corresponding title 404 will be broadcast.
The start time 401 is the time (hour:minute:second) when the broadcast program of the corresponding title 404 will be broadcast.
The duration 402 is the program duration (hours:minutes:seconds) of the broadcast program of the corresponding title 404 .

The genre identifier 403 is an identifier indicating the genre of the broadcast program of the corresponding title 404. The genre identifies the content of the broadcast program by field, such as news, sports, and drama. Here, the genre identifier is composed of higher-level categories that classify higher-level items, and lower-level categories that subdivide the higher-level categories. For example, if the higher-level category is "sports," the lower-level categories are "baseball,""soccer," and so on. Here, the genre identifier 403 is expressed as a two-digit hexadecimal number, with the higher digits representing the higher-level category and the lower digits representing the lower-level category.
As shown in FIG. 4, a plurality of genre identifiers may be set for one broadcast program.

The title 404 is the name of the broadcast program. If the broadcast program is a subtitled broadcast, predetermined characters (here, [字]) are added to the title 404. If the broadcast program is a rebroadcast, predetermined characters (here, [回]) are added to the title 404.
For example, this indicates that a program with the title “Second Time in Thailand: Bangkok Edition” [subtitles]” with a duration of 30 minutes (00:30:00) and genre identifiers of “0x25”, “0xa0”, and “0x86” will be broadcast at a start time (06:00:00) on the date (2020/07/08).

Now, with reference to FIG. 5, the genre identifier will be described in more detail.
The genre identifier can use the standard (ARIB STD-B10) defined by ARIB.
FIG. 5 shows the genre classification items defined in ARIB STD-B10 Part 2, Appendix H.

The major genre classification 500 indicates a higher-level classification of genres, and the medium genre classification 501 indicates a lower level classification obtained by further classifying the major genre classification 500. Here, in accordance with the terminology of ARIB, the higher level classification is called the major genre classification, and the lower level classification is called the medium genre classification.
The major genre classification 500 is a two-digit hexadecimal number with the most significant digits defined in advance, and the intermediate genre classification 501 is a two-digit hexadecimal number with the most significant digits defined in advance.

For example, the value "0x0" of the genre major classification 500, which is the value of the upper digits, indicates "news/reporting," and "0x1" indicates "sports."
In addition, the value "0x0" of the intermediate genre category 501, which is the lower digit value of the value "0x2" of "information/variety show" in the major genre category 500, indicates "entertainment/variety show", and "0x1" indicates "fashion".

The genre identifier is a 2-digit (8-bit) hexadecimal value with the major genre classification 500 as the upper digits (upper 4 bits) and the intermediate genre classification 501 as the lower digits (lower 4 bits).
For example, the genre identifier "0x21" indicates the genre of "information/variety show" + "fashion."
Returning to FIG. 1, the description of the configuration of the genre-specific text collection device 1 will be continued.

The EPG information extraction means 111 writes and stores the extracted EPG information in the storage means 12. In the case of a TS signal corresponding to multiple channels, the EPG information extraction means 111 extracts EPG information from each TS signal and stores it in the storage means 12. When extracting EPG information for multiple channels, the EPG information extraction means 111 may store the information in separate storage areas for each channel in the storage means 12, or may store the information without distinguishing between channels.

The subtitle information extraction means 110 and the EPG information extraction means 111 can each extract the information using existing methods. For example, ariblib (reference: https://pypi.org/project/ariblib/), a library for the programming language python, can be used.

After extracting the subtitle information and EPG information for the specified time period, the broadcast information extraction means 11 notifies the program information identification means 13 that the extraction is complete (extraction completion notification). Note that this notification is not essential if the user instructs the program information identification means 13 to do so.

The storage means 12 stores the subtitle information (see FIG. 3) and the EPG information (see FIG. 4) extracted by the broadcast information extraction means 11 .
This storage means 12 can be configured with a general storage medium such as a hard disk or a semiconductor memory.

The program information identifying means 13 identifies the time information and genre of a broadcast program from the EPG information. The program information identifying means 13 operates when an extraction completion notification is received from the broadcast information extracting means 11 or when an instruction is received from the user.
The program information identification means 13 identifies the genre of each broadcast program from the EPG information (see Figure 4) stored in the storage means 12, and also identifies the time information of the broadcast program (here, date, start time, and duration).

The program information identification means 13 identifies the genre of a broadcast program using a preset classification criterion. The classification criterion is whether to classify the genre into a major genre classification (higher classification) or into a medium genre classification (lower classification).

When the set classification criterion is a major genre classification (higher classification), the program information specification means 13 sets the value of the major genre classification (higher digits) of the genre identifier as the genre value.
Furthermore, when the set classification criterion is a genre medium classification (sub-classification), the program information specification means 13 sets the genre identifier itself as the genre value.
Below, a specific description will be given of the case where the classification criterion is a major genre classification and the case where the classification criterion is a medium genre classification.

(When the classification criteria is genre major classification [higher classification])
When the classification criterion is a major genre classification (higher classification), the program information identification means 13 deletes the lower digits (lower 4 bits), which are the genre medium classification of the genre identifier set in the broadcast program, and uses the upper digits (higher 4 bits) of the major genre classification as the identifier.
If there is one such identifier in a broadcast program, the program information specification means 13 determines that identifier as the genre of the broadcast program.

On the other hand, if multiple genre identifiers are set for one broadcast program and multiple identifiers with the highest digits (highest 4 bits) of the same major genre classification exist, the program information identification means 13 accumulates the frequency of appearance of the identifiers and determines the identifier with the highest frequency as the genre of the broadcast program. Note that if multiple identifiers with the highest frequency exist, the program information identification means 13 determines the identifier that appears first in the order of genre identifiers set in the EPG information as the genre of the broadcast program.
This makes it possible to identify the main genre of a broadcast program even if multiple genre identifiers are set for the broadcast program.

However, if there are multiple identifiers with the highest frequency, the program information identification means 13 may associate the time information of the same broadcast program with each of the multiple identifiers' genres, rather than identifying a single genre.

(If the classification criteria is genre classification [sub-classification])
When the classification criterion is a medium genre classification (sub-classification), the program information specification means 13 determines the genre identifier set in the broadcast program as the genre of the broadcast program.
When multiple genre identifiers are set for one broadcast program, the program information identification means 13 determines the genre identifier that appears first in the order of the genre identifiers set in the EPG information as the genre of the broadcast program.

However, when multiple genre identifiers are set for one broadcast program, the program information identification means 13 may associate the time information of the same broadcast program with each of the multiple genres, rather than identifying a single genre.
In this way, the program information identifying means 13 identifies the genre corresponding to the broadcast program from the EPG information, and outputs the genre and time information of the broadcast program to the text extracting means 14 .

When the program information identifying means 13 determines that a broadcast program is a rebroadcast in the EPG information, it may exclude the broadcast program from targets for identifying time information and genre. For example, the program information identifying means 13 can refer to the EPG information (FIG. 4) and determine whether a broadcast program is a rebroadcast or not based on whether a predetermined character indicating a rebroadcast (here, [re]) is set in the title 404.
In this way, by excluding reruns of broadcast programs from the targets for text extraction, it is possible to prevent duplicate text extraction for the same genre.

The program information identification means 13 may identify the time information and genre of a broadcast program only when it has determined that the broadcast program is a subtitled broadcast in the EPG information. Note that even if the time information and genre are identified for a broadcast program that is not a subtitled broadcast, there is no effect on subsequent processing, since the subtitle information simply does not exist.

The text extraction means 14 extracts subtitle text for a time period specified by time information of a broadcast program from the subtitle information, and outputs the extracted text as genre-specific text in association with the genre of the broadcast program.
As shown in FIG. 1, the text extraction unit 14 includes a subtitle text extraction unit 140 and a formatting unit 141 .

The subtitle text extraction means 140 extracts subtitle text corresponding to time information of a broadcast program from the subtitle information for each genre of the broadcast program identified by the program information identification means 13 .
A subtitle text extraction means 140 extracts subtitle text for a time section specified by time information of a broadcast program for each genre of the broadcast program from the subtitle information stored in the storage means 12 based on the program information specified by the program information specification means 13.

For example, suppose that subtitle information such as that shown in Figure 6 is stored in the storage means 12, and the program information identification means 13 notifies the program of 0x02 (major genre classification [higher classification]) as the genre, and the date (2020/07/08), start time (06:00:00), and duration (00:30:00) as the time information.
In this case, the subtitle text extraction unit 140 extracts the subtitle text 302 for 30 minutes from 6:00 on July 8, 2020, as the subtitle text corresponding to the genre 0x02 (information/variety show). The same applies to the other genre 0x08 (documentary/educational).

In this way, the subtitle information stored in the storage means 12 is associated with the subtitle text 302 in association with the time information (date 300, start time 301), so the subtitle text extraction means 140 can extract the subtitle text corresponding to the genre from the genre and time information notified by the program information identification means 13.
The subtitle text extraction means 140 outputs the extracted subtitle text for each genre to the shaping means 141 .

The shaping means 141 converts the subtitle text into shaped text by deleting text (meta information) other than the spoken text from the genre-specific subtitle text extracted by the subtitle text extraction means 140.

The meta-information used in subtitles is collected into certain patterns, such as speaker notation and scene notation. Therefore, the shaping means 141 can shape the subtitle text by performing a predetermined regular expression filter process.

An example of the regular expression filter process of the formatter 141 will be described with reference to FIG.
7A shows an example of deleting a speaker notation. For example, in the case of "speaker" + ">>", such as "Ana>>", the shaping unit 141 determines the text from the beginning of the sentence to just before the ">>" as the speaker, and deletes the ">>" from the beginning of the sentence.
7B shows an example of deleting a scene description. For example, for a scene description character string enclosed in parentheses, such as (applause and cheers), the shaping unit 141 deletes the scene description character string together with the parentheses.

7C is an example of connecting segmented phrases. In the case of subtitles, a sentence of subtitles is divided depending on the scene, and for example, [⇒] is used to indicate that the next segment continues. In this case, the shaping unit 141 connects the segments by replacing "[⇒]" with a comma ",".
7D shows an example of deleting background sound notation. For example, in subtitles, a symbol 600 indicating the sound of a telephone ringing, a symbol 601 indicating the sound of someone talking in the background, etc. are used as background sounds. In this case, the shaping means 141 deletes the symbols 600 and 601 indicating the background sounds.
This enables the shaping unit 141 to convert the subtitle text into text that contains only the spoken content, without notations specific to subtitles.
Returning to FIG. 1, the description of the configuration of the genre-specific text collection device 1 will be continued.

The shaping means 141 outputs the text shaped by genre as genre-specific text. The output destination of the shaping means 141 is a storage device (not shown) that is directly connected or connected via a network.

With the above-described configuration, the genre-specific text collection device 1 can collect genre-specific text from subtitle text based on EPG information simply by receiving broadcast waves.
In addition, since genres are defined based on broadcasting standards, differences between broadcasting stations are unlikely to occur, and the genre-specific text collection device 1 can collect large amounts of high-quality text corpora for each genre that can be used for learning language models and natural language processing.

The genre-specific text collection device 1 can be operated using a program (genre-specific text collection program) that causes a computer to function as each of the above-mentioned means.

<Operation of the genre-specific text collection device>
Next, the operation of the genre-specific text collection device 1 according to the embodiment of the present invention will be described with reference to FIGS. 8 and 9 (and for the configuration, refer to FIG. 1 as appropriate).

(Overall operation)
First, the overall operation of the genre-specific text collection device 1 will be described with reference to FIG.
In step S1, the broadcast receiving means 10 receives a digital broadcast wave and demodulates it into a TS signal. At this time, the broadcast receiving means 10 demodulates a broadcast signal of a specified channel, and the number of channels may be one or more.

In step S2, the subtitle information extraction means 110 of the broadcast information extraction means 11 analyzes the TS signal demodulated in step S1 and extracts subtitle information from the TS signal for a specified time period. Here, the subtitle information extraction means 110 extracts date 300, start time 301, and subtitle text 302 from the TS signal, as shown in FIG. 3, and stores them in the storage means 12.

In step S3, the EPG information extraction means 111 of the broadcast information extraction means 11 analyzes the TS signal demodulated in step S1 and extracts EPG information from the TS signal for a specified time period. Here, the EPG information extraction means 111 extracts date 400, start time 401, duration 402, genre identifier 403 and title 404 from the TS signal, as shown in Fig. 4, and stores them in the storage means 12.
It should be noted that steps S2 and S3 do not necessarily have to be performed in this order, and steps S3 and S2 may be performed in this order, or steps S2 and S3 may be performed in parallel.

In step S4, the broadcast information extraction means 11 determines whether or not the designated time period has been completed.
If the specified time period has not yet been completed (No in step S4), the genre-specific text collection device 1 returns to step S2 and continues the operation.

On the other hand, if the specified time period has been completed (Yes in step S4), in step S5, the program information identification means 13 reads out the EPG information for each broadcast program from the EPG information stored in the storage means 12 in step S3.

In step S6, the program information identification means 13 identifies the genre of the broadcast program using preset classification criteria. The detailed operation of identifying the genre in step S6 will be described later (see FIG. 9).

In step S7, the subtitle text extraction means 140 of the text extraction means 14 extracts subtitle text corresponding to the time information (here, date, start time, and duration) included in the EPG information for each broadcast program read in step S5 from the subtitle information stored in the storage means 12 in step S2.

In step S8, the shaping means 141 of the text extraction means 14 shapes the subtitle text extracted in step S7 by performing a predetermined regular expression filter process.
In step S9, the shaping means 141 outputs the shaped text together with the genre identified in step S6 as genre-specific text.

In step S10, the program information specifying means 13 judges whether or not there is EPG information of a broadcast program in the storage means 12 that has not yet been read out.
If there is EPG information of a broadcast program that has not yet been read out (Yes in step S10), the genre-specific text collection device 1 returns to step S5 and continues the operation.
On the other hand, if all the EPG information has been read out (No in step S10), the genre-specific text collection device 1 ends the operation.

(Genre specific action)
Next, the operation of step S6 (FIG. 8) performed by the program information specifying means 13 will be described in more detail with reference to FIG.

In step S61, the program information specifying means 13 determines whether to classify the genre into a major genre classification (higher classification) or into a medium genre classification (lower classification) based on a preset classification criterion.
If classification is to be made based on the genre medium classification (sub-classification) (No in step S61), the program information specification means 13 advances the operation to step S63.

On the other hand, when classifying by major genre classification (higher classification) (Yes in step S61), in step S62, the program information specification means 13 deletes the lowest digits of the genre identifier.
In step S63, the program information specifying means 13 judges whether or not one genre identifier is set for the broadcast program.

If there is only one genre identifier (Yes in step S63), in step S64, the program information identification means 13 identifies that genre identifier (the upper digits in the case of a major genre classification) as the genre of the broadcast program.

On the other hand, if there are multiple genre identifiers (No in step S63), in step S65, the program information identification means 13 accumulates the occurrence frequencies of the multiple genre identifiers (higher digits in the case of major genre classifications).
In step S66, the program information specifying means 13 judges whether the genre identifier (the upper digits in the case of a major genre classification) having the maximum frequency accumulated in step S65 is one or not.
If there is one genre identifier with the highest frequency (Yes in step S66), in step S67, the program information identification means 13 identifies the genre identifier with the highest frequency (the most significant digits in the case of a major genre classification) as the genre of the broadcast program.

On the other hand, if there are multiple genre identifiers with the highest frequency (No in step S66), in step S68, the program information identification means 13 identifies the genre identifier (the most significant digits in the case of a major genre classification) that appears first in the EPG information set for the broadcast program as the genre of the broadcast program.

After steps S64, S67 and S68, the program information identifying means 13 ends the operation of step S6 and proceeds to step S7 (FIG. 8).
By the above operation, the genre-specific text collection device 1 can collect genre-specific text from subtitle text based on EPG information, simply by receiving broadcast waves.

REFERENCE SIGNS LIST 1 Genre-specific text collection device 10 Broadcast receiving means 11 Broadcast information extraction means 110 Subtitle information extraction means 111 EPG information extraction means 12 Storage means 13 Program information identification means 14 Text extraction means 140 Subtitle text extraction means 141 Formatting means

Claims (6)

A genre-specific text collection device that collects genre-specific text from subtitle text multiplexed in digital broadcasting, comprising:
A broadcast receiving means for receiving and demodulating the digital broadcast;
a subtitle information extraction means for extracting subtitle information including the subtitle text and time information for presenting the subtitle text from the signal demodulated by the broadcast receiving means;
an EPG information extraction means for extracting EPG information of a broadcast program from the demodulated signal;
a program information specifying means for specifying time information and a genre of the broadcast program from the EPG information;
a text extraction means for extracting from the caption information a caption text for a time period specified by time information of the broadcast program, and associating the extracted caption text with a genre of the broadcast program to generate a genre-specific text ,
The genres are composed of higher-level categories classified by higher-level items and lower-level categories subdivided from the higher-level categories, and one or more genres are set for each of the broadcast programs in the EPG information,
The program information identification means identifies only the higher-level classification as the genre, and if multiple genres are set in the EPG information, identifies the higher-level classification that is most frequently set in the broadcast program as the genre . The genre-specific text collection device according to claim 1, characterized in that, when the number of highest-ranking categories set for the broadcast programs is the same, the program information identification means identifies the highest-ranking category that appears first in the EPG information as the genre. 3. The genre-specific text collection device according to claim 1, wherein the text extraction means deletes meta-information other than spoken text from the subtitle text. The genre-specific text collection device according to any one of claims 1 to 3, characterized in that if the program information identification means determines that the broadcast program is a rebroadcast in the EPG information, it excludes the broadcast program from being targeted for identifying time information and genre. The genre-specific text collection device according to any one of claims 1 to 4 , characterized in that the program information identification means identifies the time information and genre of the broadcast program only when it is determined in the EPG information that the broadcast program is a subtitled broadcast. A genre-specific text collection program for causing a computer to function as the genre-specific text collection device according to any one of claims 1 to 5 .

Images