JP2014203383A - Relay section extraction device, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To extract relay sections contained in a program.SOLUTION: A location identification unit 11 of a relay section extraction device 1 extracts expressions related to a location from respective sentences represented by text data associated with video data of a news program and identifies the location from the extracted expressions. A sentence classification unit 12 determines whether or not the respective sentences represented by the text data are comments related to a relay. On the basis of types of respective comments of two different sentences that have been determined as comments using expressions related to the relay by the sentence classification unit 12 and a difference between locations of the two different sentences that the location identification unit 11 has identified, a relay section extraction unit 15 extracts sections in the video data corresponding to the two different sentences, as the relay sections.

Description

  The present invention relates to a relay section extraction device and a program.

  Conventionally, as a technique to classify scenes such as soccer broadcasts, a method of classifying soccer broadcast announcement comments into two types of game description and commentary, and extracting a set of match description sentences as one meaningful section, A method for extracting a section in which a characteristic expression in a program is used has been proposed (see, for example, Non-Patent Document 1 and Non-Patent Document 2).

Ichiro Yamada, Masanori Sano, Hideki Sumiyoshi, Masahiro Shibata, Nobuyuki Yagi, “Automatic Generation of Soccer Program Segment Metadata Using Comments of Announcer and Commentator”, IEICE Transactions D, Vol. J89-D, No. 10, 2006, p.2328-2337 Ichiro Yamada, Kikuka Miura, Yoshihiko Kawai, Hideki Sumiyoshi, Nobuyuki Yagi, Manabu Okumura, Kennobu Tokunaga, “Extraction of typical sentence sections in closed captions using similarity in global sentence structure”, electronic information communication Academic Journal D, Vol. J90-D, No. 9, 2007, p. 2624-2633

  In the case of soccer broadcast, the announcement comments in the section where the event occurs have a similar tendency. However, in the case of news programs, the number of vocabulary types used for comments in the relay section is large, and not all are similar. Therefore, it is difficult to extract a relay section from a program that conveys various topics, such as a news program, even if a conventional method for extracting a section in which a characteristic expression is used in the program is applied.

  The present invention has been made in consideration of such circumstances, and provides a relay section extraction apparatus and program that can extract a relay section included in a program.

[1] According to one aspect of the present invention, a place specifying unit that extracts a phrase related to a place from each sentence indicated by text data associated with video data of a program and specifies the place from the extracted phrase, the text A sentence classification unit that determines whether each sentence indicated by the data is a comment that uses an expression related to relay, and two different sentences that are determined to be comments that use an expression related to relay by the sentence classification part The section of the video data corresponding to the two different sentences is extracted as a relay section based on the type of the comment and whether the place specified by the place specifying unit is different for each of the two different sentences. And a relay section extraction unit.
According to this invention, the relay section extraction device specifies a location for each sentence indicated by the text data associated with the video data of the program, and determines whether or not the comment is a comment using an expression related to relay. . The relay section extraction device determines whether the video data is based on the type of comment for each of the two sentences determined to be comments using expressions related to the relay and whether or not the locations specified for the sentences are different. Extract relay section.
Thereby, the relay section extraction device can extract the scene of the relay section included in the video of the news program by using the caption broadcast of the news program.

[2] One aspect of the present invention is the above-described relay section extraction device, wherein the sentence classification unit determines whether each sentence indicated by the text data is a relay entry comment or not. It is determined whether or not there is a relay closing comment, the relay section extraction unit, the location specified by the location specifying unit for the sentence determined to be a relay closing comment, and If there is no difference in the location specified by the location specifying unit for other sentences that appear before the sentence in the text data and are determined to be relay comments, or in the text data If a sentence determined to be a relayed comment appears before a sentence determined to be a closing comment, it is determined to be a relayed comment or a relayed comment. The section of the video data corresponding to the sentence determined to be a relay closing comment from the other sentence is used as a relay section by a local reporter.
According to the present invention, the relay section extraction device is specified for the location specified for the sentence determined as the relay closing comment, and the other sentence that appears before the sentence and is determined as the relay comment. If there is no difference in location, video data sections corresponding to these sentences are extracted as relay sections by local reporters. In addition, when a sentence that is determined to be a relay comment appears before a sentence that is determined to be a relay closing comment, the relay section extraction device sets a section of video data corresponding to these sentences to the local This is extracted as a relay section by the reporter.
Thereby, the relay section extraction device can extract the relay section by the local reporter in the news program using the subtitle broadcast data of the news program.

[3] One aspect of the present invention is the relay section extraction device described above, wherein the sentence classification unit further determines whether each sentence indicated by the text data is a comment that conveys a relay location. The relay section extraction unit is a comment that tells the place that the place specifying unit has specified for the sentence determined to be a relay swing comment, and that appears after the sentence in the text data and that tells the place If there is no difference between the place specified by the place specifying unit for the other sentence determined, from the sentence determined to be a relay swing comment to the other sentence determined to be a comment conveying the place The section of the video data corresponding to is extracted as a relay section by a local reporter.
According to this invention, the relay section extraction device is specified for a place specified for a sentence determined as a relay swing comment, and for another sentence that appears after the sentence and is determined as a comment that conveys the place. If there is no difference in location, video data sections corresponding to these sentences are extracted as relay sections by local reporters.
Thereby, the relay section extraction device can extract the relay section by the local reporter in the news program even when there is no relay closing comment using the caption broadcast data of the news program.

[4] One aspect of the present invention is the relay section extraction device described above, wherein the sentence classification unit determines whether each sentence indicated by the text data is a comment that conveys a relay location, and relays. Whether or not the comment is a comment explaining the video, and the relay section extraction unit includes the location specified by the location specifying unit for the sentence determined to be a comment conveying the location of the relay, and the text data If there is no difference between the place specified by the place specifying unit with respect to another sentence that appears after the sentence and is determined to be a comment that conveys the location of the relay or a comment that describes the video of the relay, Before the sentence determined to be a comment to be communicated, and the other sentence determined to be a comment that conveys the location of the relay or a comment that describes the video of the relay A section of the video data is extracted as a relay section where the reporter is absent from the site.
According to the present invention, the relay section extracting device displays a place specified for a sentence that is determined to be a comment that tells a relay location, and a comment or a relay video that appears after the sentence and tells the relay location. If there is no difference between the comment to be explained and the locations specified for the other sentences determined to be, the reporter extracts sections of the video data corresponding to these sentences as relay sections that are not present in the field.
Thereby, the relay section extraction device can extract the relay section where the reporter in the news program is absent from the site by using the caption broadcast data of the news program.

[5] One aspect of the present invention is the above-described relay section extraction device, wherein the relay section extraction unit specifies the location for each of two different sentences that are determined to be comments describing a video of relay. If there is no difference in the location specified by the section, the section of the video data corresponding to the two different sentences is extracted as a relay section that is absent from the site.
According to the present invention, when there is no difference in the location specified for each of two sentences determined to be comments explaining the video of the relay, the relay section extraction device selects the section of the video data corresponding to these sentences as the reporter. Is extracted as a relay section that is absent from the site.
As a result, the relay section extraction device can extract the relay section where the reporter in the news program is absent from the site even when there is no comment specifying the location, using the caption broadcast data of the news program. .

[6] One aspect of the present invention is the above-described relay section extraction device, wherein a sentence in the text data corresponding to the relay section extracted by the relay section extraction unit includes a phrase representing aerial photography. A relay section extracting unit that determines that the relay section is a relay section by aerial photography.
According to this invention, the relay section extraction device determines whether the relay section is a relay section by aerial photography based on the sentence indicated by the text data of the relay section extracted by the relay section extraction unit.
Thereby, the relay section extraction device can extract the relay section by the aerial shooting from the relay section by the local reporter in the program and the relay section by which the reporter is absent from the site.

[7] According to one aspect of the present invention, a computer used as a relay section extracting device extracts a phrase related to a place from each sentence indicated by text data associated with video data of a program, and a place from the extracted phrase A location identifying unit that identifies a sentence, a sentence classification unit that determines whether each sentence indicated by the text data is a comment that uses an expression related to relay, and a comment that uses an expression related to relay by the sentence classification unit The video corresponding to the two different sentences based on the type of the comment for each of the two different sentences determined and whether the location specified by the place specifying unit is different for each of the two different sentences. This is a program for causing a data section to function as a relay section extracting unit that extracts a section as a relay section.

  According to the present invention, a relay section included in a program can be extracted.

It is a functional block diagram which shows the structure of the relay area extraction apparatus in one Embodiment of this invention. It is a figure which shows the example of the news text of the embodiment. It is a figure which shows the example of the sentence analysis data of the embodiment. It is a figure which shows the example of the relay area data of the embodiment. It is a figure which shows the flowchart of the relay area extraction process by the relay area extraction apparatus of the embodiment. It is a figure which shows the flowchart of the place specific process by the place specific part of the embodiment. It is a figure which shows the flowchart of the 1st sentence classification | category process by the 1st sentence classification | category part of the embodiment. It is a figure which shows the flowchart of the 2nd sentence classification | category process by the 2nd sentence classification | category part of the embodiment. It is a figure which shows the flowchart of the local reporter relay area extraction process by the local reporter relay area extraction part of the embodiment. It is a figure which shows the flowchart of the reporter absence relay area extraction process by the reporter absence relay area extraction part of the embodiment.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

[Overview]
For news programs broadcast on TV, etc., in addition to the scene explained by the announcer in the studio, the reporter explains from the site where it is relayed, the reporter is absent from the site, the relay by aerial photography from a helicopter, etc. Scenes are included. There are relays where the reporter is absent from the site, such as relays where people take pictures with a camera, as well as relays where they take pictures with a robot camera. In the present embodiment, the relay scenes are extracted and classified from the news program by using text data such as caption broadcasting accompanying the news program. As a result, it is possible to acquire information as to which news program was reported from which location and in what form.

[Configuration of Relay Section Extraction Device]
FIG. 1 is a functional block diagram showing a configuration of a relay section extraction device 1 according to an embodiment of the present invention. The relay section extraction device 1 is realized by, for example, one or a plurality of computer devices. As shown in the figure, the storage section 10, the location identification section 11, the sentence classification section 12, the relay section extraction section 15, and noise A removal unit 19 is provided. News text is input to the relay section extraction device 1. The news text is text data associated with the video (moving image) data of the news program, and may be subtitle broadcast accompanying the news program or may be a result of voice recognition of the news program. Note that the news program may include only one topic (news item) or may include a plurality of topics.

  The storage unit 10 stores various data such as news text, sentence analysis data, and relay section data. The sentence analysis data indicates the sentence analysis result of each sentence indicated by the news text. The relay section data indicates the form of relay, the start position and end position of the relay section in the program, and the relay location. As for the form of the relay, the relay explained by the reporter from the field (hereinafter referred to as “local reporter relay”), the relay where the reporter is not present in the field (hereinafter referred to as “reporter absent relay”), aerial photography (Hereinafter referred to as “aerial shooting relay”).

  The location specifying unit 11 extracts a phrase related to the location from each sentence indicated by the news text, and specifies the location from the extracted phrase. In the present embodiment, the place specifying unit 11 acquires a place name such as a city name as information specifying the place from a place name or a landmark name that is a word related to the place. In addition, you may use the code | cord | chord etc. which specify a municipality as information which specifies a place. The place identifying unit 11 writes sentence analysis data indicating the place name obtained for each sentence in the storage unit 10.

  The sentence classification unit 12 includes a first sentence classification unit 13 and a second sentence classification unit 14, determines whether each sentence indicated by the news text is a comment using an expression related to relay, and stores the determination result Write to the sentence analysis data stored in the unit 10. The first sentence classification unit 13 determines whether each sentence indicated by the news text is a comment using an expression for specifying a place, or a comment using an expression for explaining a video. The second sentence classifying unit 14 determines whether each sentence indicated by the news text is a relay comment from a local relay, whether it is a comment from a relay from the local, It is determined whether or not.

  The relay section extraction unit 15 is based on the type of comment for each of the two sentences determined to be comments using expressions related to the relay, and whether or not the location specified for each of the two sentences is different. Then, the relay section is extracted. The relay section extraction unit 15 writes the relay section data in which the relay form, start position and end position, and relay location of the extracted relay section are set in the storage unit 10. The relay section extraction unit 15 includes a local reporter relay section extraction unit 16, a reporter absent relay section extraction unit 17, and an aerial shooting relay section extraction unit 18. The local reporter relay section extraction unit 16 determines the local reporter relay section and the relay location based on the determination result by the first sentence classification unit 13 and the second sentence classification unit 14 and the place specified by the place specifying unit 11. Extract. The reporter absent relay section extraction unit 17 is based on the extraction result by the local reporter relay section extraction unit 16, the determination result by the first sentence classification unit 13, and the location specified by the place specifying unit 11. And the location of the relay. The aerial shooting relay section extraction unit 18 extracts the aerial shooting relay section from the local reporter relay section extracted by the local reporter relay section extraction unit 16 and the reporter absence relay section extracted by the reporter absence relay section extraction unit 17. To do. The noise removal unit 19 includes a local reporter relay segment extracted by the local reporter relay segment extraction unit 16, a reporter absent relay segment extracted by the reporter absent relay segment extraction unit 17, and an aerial relay segment extracted by the aerial broadcast segment extraction unit 18. The noise that is not actually a relay section is excluded from the list.

  FIG. 2 is a diagram illustrating an example of news text. As shown in the figure, news text is text data indicating a sentence corresponding to video data of a news program, and synchronization data indicating association with video data is added. The synchronization data is indicated by, for example, the reproduction time from the beginning of the video data.

  FIG. 3 is a diagram showing sentence analysis data stored in the storage unit 10. As shown in the figure, each record constituting the sentence analysis data indicates a sentence number, a location specified from the sentence having the sentence number, and a kind of sentence having the sentence number. The sentence number i indicates the i-th sentence that appears in the news text. The city name is set in the place. The type of sentence is represented by the values of the function Kind1S (i) and the function Kind2S (i). The determination result by the first sentence classification unit 13 is set in Kind1S (i), and the determination result by the second sentence classification unit 14 is set in Kind2S (i).

  FIG. 4 is a diagram illustrating relay section data stored in the storage unit 10. As shown in the figure, each record constituting the relay section data indicates the form of relay, the start position and end position of the relay section, and the relay location. The form of the relay is a local reporter relay, a reporter absent relay, or an aerial shooting relay. The start position and end position of the relay section are indicated by sentence numbers, and the video data is relayed from the reproduction time indicated by the synchronization data corresponding to the sentence number at the start position to the reproduction time indicated by the synchronization data corresponding to the sentence number at the end position. It is a section. Note that the reproduction time indicated by the synchronization data corresponding to the sentence at the start position and the sentence at the end position may be set as the start position and end position of the relay section.

[Operation of relay section extractor]
Subsequently, the operation of the relay section extraction device 1 will be described.
FIG. 5 is a flowchart showing a processing procedure of relay section extraction processing by the relay section extraction apparatus 1 shown in FIG. First, a subtitle broadcast accompanying a news program is input to the relay section extracting device 1 as news text. The news text indicates n sentences (n is an integer of 1 or more), and the i-th sentence (i is an integer of 1 to n) indicated by the news text is referred to as a sentence S (i). The storage unit 10 stores the input news text.

<Step S1-1: Location Identification Process>
The place specifying unit 11 extracts a place name and a landmark name from each sentence indicated by the input news text, and specifies a city name. This detailed process will be described with reference to FIG.

FIG. 6 is a flowchart showing the processing procedure of the location specifying process by the location specifying unit 11, and shows the detailed processing procedure of step S1-1 in FIG.
First, the place identifying unit 11 substitutes 1 for i in order to set the sentence to be processed as a sentence head (step S2-1). Next, the place identifying unit 11 extracts a place name and a landmark name (landmark name or name) from the i-th sentence S (i) to be processed (step S2-2). For this extraction process, for example, an existing unique name extractor (such as cabocha shown in “http://code.google.com/p/cabocha/”) can be used.

  Subsequently, the place specifying unit 11 specifies a municipality from the place name and landmark name extracted in step S2-2 (step S2-3). In order to specify the municipality, for example, the name of the municipality is stored in advance in the storage unit 10 in association with the place name or landmark name of the whole country. The location specifying unit 11 reads out the city name associated with the place name or landmark name extracted in step S2-2 from the storage unit 10. Thereby, for example, when the place identifying unit 11 extracts the place name “Kesennuma” in Step S2-2, “Kesennuma” is identified as “Kesennuma City, Miyagi Prefecture” in Step S2-3.

  Note that there may be a plurality of municipalities corresponding to the place names and landmark names extracted in step S2-2. In this case, the place specifying unit 11 sets them as candidates for municipalities. The location identifying unit 11 identifies a municipality from among candidates using a conventional method based on the number of co-occurrence in a sentence. For example, when the place name “Ofunato” is extracted in step S2-2, the three city areas of “Ofunato, Kanayama-cho, Gero-shi, Gifu”, “Ofunato-cho, Ofunato-shi, Iwate” and “Ofunato-cho, Sakai-shi, Niigata” Towns and villages are candidates. When the place name is included in the current sentence S (i) to be processed, the place specifying unit 11 selects a municipality according to the inclusion relation with the place name and the co-occurrence frequency from among the candidates for the three municipalities. Identify the town and village. For example, if the current processing target sentence S (i) has a word with a place name of “Iwate Prefecture”, the place identification unit 11 includes “Iwate Prefecture” from among the candidates. As a city. Or, for example, if the current processing target sentence S (i) has a word with a place name of “Tohoku”, the location identification unit 11 “Iwate Prefecture” has a high co-occurrence frequency with “Tohoku” among the candidates. "Ofunato-cho, Ofunato-shi, Iwate" is specified as a municipality. In this way, the place specifying unit 11 specifies one city from among the candidates using the place name appearing in the same sentence as a clue. Since sentences with similar appearance order are highly related, if the place name is not included in the same sentence, use the place names that appear in the sentences before and after the predetermined range from the sentence to be processed. Identify the city from the inside. At this time, a place name that appears in a sentence before and after a predetermined range may be identified from candidates using the place name that appears in the sentence closest to the sentence to be processed among the sentences in which the place name appears. Of these candidates, the city may be identified from the candidates using the place names with the largest number of co-occurrence.

  Moreover, when the landmark name is extracted in step S2-2, the place specifying unit 11 may specify a city / town using an existing map search service. For example, “Odaiba” is a name rather than an address. The location specifying unit 11 acquires latitude / longitude information of the extracted landmark name “Odaiba” by using a map search service (for example, Google map API) provided on the web, and the acquired latitude The municipality is identified as “Minato-ku, Tokyo” from the longitude information. In order to identify the city from the latitude / longitude information, a method for acquiring the city / town name from the latitude / longitude information again using an existing map search service (for example, Google map API) or the storage unit 10 in advance. For example, a method of retrieving latitude and longitude information of registered municipalities in the whole country can be used.

  If the sentence analysis data has not yet been generated, the location specifying unit 11 newly generates the sentence analysis data and registers it in the storage unit 10. The place specifying unit 11 adds a record of sentence number i to the sentence analysis data, and writes the city name as place information. However, when the municipality is not specified, the place specifying unit 11 keeps the place information as NULL.

  The location specifying unit 11 determines whether all sentences included in the news text have been processed by determining whether i is smaller than n (step S2-4). If the location identifying unit 11 determines that i is smaller than n and there is an unprocessed sentence (step S2-4: YES), 1 is added to i and the sentence to be processed is shifted backward by one. (Step S2-5), the processing from Step S2-2 is repeated. And the place specific | specification part 11 complete | finishes the process of step S1-1 of FIG. 5, when it determines with i having reached n and having completed the process about all the sentences which a news text shows (step S2-4: NO). To do.

<Step S1-2: First sentence classification process>
In step S1-2 of FIG. 5, the first sentence classification unit 13 determines whether each sentence indicated by the news text is a comment specifying a place or a comment explaining a video. This detailed process will be described with reference to FIG.

FIG. 7 is a flowchart showing a processing procedure of the first sentence classification processing by the first sentence classification unit 13, and shows a detailed processing procedure of step S1-2 in FIG.
First, the first sentence classification unit 13 substitutes 1 for i in order to set a sentence to be processed as a sentence head (step S3-1). Next, the first sentence classification unit 13 determines whether or not the i-th sentence S (i) to be processed is a comment specifying a place (step S3-2). The comment specifying the location is a comment that directly conveys the location information about the video, such as “current Kesennuma” or “prefectural office video”. Therefore, a comment specifying the location and a comment not specifying the location are prepared as learning data, and the features of the comment specifying the location are learned in advance by machine learning such as SVM (support vector machine). The first sentence classification unit 13 uses this learning result to determine whether or not the sentence S (i) is a comment specifying a place. As features, information such as words appearing in sentences and their dependencies can be used. For example, if it is “current Kesennuma state”, it depends on words such as “present”, “Kesenuma”, “no”, “situation”, “is”, and “situation” of “Kesenuma”. Is determined to be a comment specifying a place.

  When the first sentence classification unit 13 determines that the sentence S (i) is a comment specifying a place (step S3-2: YES), the function Kind1 (S (i)) indicating the type of the sentence S (i) ) Is set to a value indicating “location specific comment” (step S3-3). That is, the first sentence classification unit 13 writes a value indicating “location specific comment” in Kind1 (S (i)) included in the record of sentence number i of the sentence analysis data. The first sentence classification unit 13 sets the value of Kind1 (S (i)), and then proceeds to the process of step S3-6.

  On the other hand, when it is determined that the sentence S (i) is not a comment specifying the place (step S3-2: NO), the first sentence classification unit 13 performs the process of step S3-4. That is, the first sentence classification unit 13 determines whether or not the i-th sentence S (i) is a comment explaining the video (step S3-4). The comment explaining the image is a comment about the phenomenon that has occurred in the image. For example, “You can see the car running on the quay now.”, “The right corner of the screen Can see the pool. " Therefore, a comment explaining the video and a comment not explaining the video are prepared as learning data, and features of the comment explaining the video are learned in advance by machine learning such as SVM. The first sentence classification unit 13 uses this learning result to determine whether or not the sentence S (i) is a comment explaining the video. As features, information such as words appearing in sentences and their dependencies can be used.

When the first sentence classification unit 13 determines that the sentence S (i) is a comment explaining the video (step S3-4: YES), Kind1 (S ( In i)), a value indicating “video comment” is set (step S3-5). The first sentence classification unit 13 sets the value of Kind1 (S (i)), and then proceeds to step S3-6.
If the first sentence classification unit 13 determines that the sentence S (i) is not a comment explaining the video (step S3-4: NO), the first sentence classification unit 13 proceeds to step S3-6.

  After the process of step S3-3 or step S3-5, or when it is determined NO in the process of step S3-4, the first sentence classification unit 13 determines whether i is smaller than n. Then, it is determined whether all sentences included in the news text have been processed (step S3-6). If the first sentence classification unit 13 determines that i is smaller than n and there is an unprocessed sentence (step S3-6: YES), 1 is added to i and the sentence to be processed is moved backward. Shift (step S3-7) and repeat the process from step S3-2. When the first sentence classification unit 13 determines that i has reached n and all the sentences indicated by the news text have been processed (step S3-6: NO), the process of step S1-2 in FIG. Exit.

<Step S1-3: Second sentence classification process>
In step S1-3 of FIG. 5, the second sentence classification unit 14 determines whether each sentence indicated by the news text is an on-the-job relay comment, whether it is an on-site relay comment, It is determined whether it is a closing comment. This detailed processing will be described with reference to FIG.

FIG. 8 is a flowchart showing the processing procedure of the second sentence classification process by the second sentence classification unit 14 and shows the detailed process of step S1-3 in FIG.
First, the second sentence classification unit 14 substitutes 1 for i in order to set the sentence to be processed as a sentence head (step S4-1). Next, the second sentence classification unit 14 determines whether or not the i-th sentence S (i) to be processed is a relay swing comment from the site (step S4-2). The comment on the broadcast from the site is a comment issued by the announcer of the studio when entering the broadcast on a news program, for example, “Now, I will relay it from the Ofunato refuge in Iwate Prefecture”, “Now, I will relay you from the south exit of Shinjuku Station in Tokyo.” Therefore, the relay comment from the site and the comment that is not so are prepared as learning data, and the features of the relay comment from the site are learned in advance by machine learning such as SVM. The second sentence classification unit 14 uses this learning result to determine whether the sentence S (i) is a relay swing comment from the site. As features, information such as words appearing in sentences and their dependencies can be used.

  If the second sentence classification unit 14 determines that the sentence S (i) is a relay comment from the field (step S4-2: YES), the function Kind2 (S ( In i)), a value indicating “relaying comment” is set (step S4-3). That is, the second sentence classification unit 14 writes a value indicating “relaying comment” in Kind2 (S (i)) included in the record of sentence number i of the sentence analysis data. The second sentence classification unit 14 sets the value of Kind2 (S (i)), and then proceeds to step S4-8.

  On the other hand, if the second sentence classification unit 14 determines that the sentence S (i) is not a relay comment from the field (step S4-2: NO), the i-th sentence S (the process target) ( It is determined whether i) is a relay comment from the field (step S4-4). The comment from the on-site broadcast is the first comment from a local announcer or reporter when entering a news program, for example, “Ofunato Daiichi Hospital on the coastal plateau in Iwate Prefecture”, “Sendai It ’s a station. ” In view of this, a comment that is relayed from the field and a comment that is not so are prepared as learning data in advance, and the features of the comment that is relayed from the field are learned by machine learning such as SVM. The second sentence classification unit 14 uses this learning result to determine whether or not the sentence S (i) is an incoming comment from the field. As features, information such as words appearing in sentences and their dependencies can be used.

  When the second sentence classification unit 14 determines that the sentence S (i) is a comment included in the relay from the field (step S4-4: YES), Kind2 (included in the record of the sentence number i of the sentence analysis data) In S (i)), a value indicating “comment with relay” is set (step S4-5). The second sentence classification unit 14 sets the value of Kind2 (S (i)), and then proceeds to step S4-8.

  On the other hand, if the second sentence classification unit 14 determines that the sentence S (i) is not a relayed comment from the field (step S4-4: NO), the i-th sentence S (the process target) ( It is determined whether i) is a relay closing comment from the site (step S4-6). The final comment on the broadcast from the site is the last comment issued by a local announcer or reporter during the broadcast in a news program. For example, “This is Ofunato Daiichi Hospital in Iwate Prefecture.”, “Shinjuku Station” I told you from the South Exit. " In view of this, a comment for closing the relay from the site and a comment that is not so are prepared as learning data, and the features of the closing comment for the relay from the site are learned in advance by machine learning such as SVM. Using the learning result, the second sentence classification unit 14 determines whether the sentence S (i) is a relay closing comment from the site. As features, information such as words appearing in sentences and their dependencies can be used.

When the second sentence classification unit 14 determines that the sentence S (i) is a closing comment from the field (step S4-6: YES), Kind2 (included in the record of the sentence number i in the sentence analysis data) In S (i)), a value indicating a “relay closing comment” is set (step S4-7). The second sentence classification unit 14 sets the value of Kind2 (S (i)), and then proceeds to step S4-8.
On the other hand, if the second sentence classification unit 14 determines that the sentence S (i) is not a closing comment for relay from the site (step S4-6: NO), the process proceeds to step S4-8.

  After the process of step S4-3, step S4-5, or step S4-7, or when it is determined NO in the process of step S4-6, the second sentence classification unit 14 determines whether i is smaller than n. By determining whether or not, it is determined whether all sentences included in the news text have been processed (step S4-8). If the second sentence classification unit 14 determines that i is smaller than n and there is an unprocessed sentence (step S4-8: YES), 1 is added to i and the sentence to be processed is moved backward. Shift (step S4-9) and repeat the process from step S4-2. When the second sentence classification unit 14 determines that i has reached n and all the sentences indicated by the news text have been processed (step S4-8: NO), the process of step S1-3 in FIG. Exit.

[Step S1-4: Local Reporter Relay Section Extraction Processing]
In step S1-4 of FIG. 5, the local reporter relay section extraction unit 16 performs the processing result of the location specifying process (step S1-1), the processing result of the first sentence classification process (step S1-2), and the second Based on the processing result of the sentence classification processing (step S1-3), the local reporter relay section and the relay location are extracted. This detailed process will be described with reference to FIG.

FIG. 9 is a flowchart showing the processing procedure of the local reporter relay section extraction processing by the local reporter relay section extraction unit 16, and shows the detailed processing procedure of step S1-4 in FIG.
First, the local reporter relay section extraction unit 16 substitutes n for i in order to set the sentence to be processed to the end of the sentence (step S5-1). Next, the local reporter relay section extraction unit 16 determines whether or not the i-th sentence S (i) to be processed is a “relay closing comment”, and the Kind2 (S It is determined by whether (i)) is a value indicating “relay closing comment” (step S5-2). When it is determined that the sentence S (i) is a “relay closing comment” (step S5-2: YES), the local reporter relay section extraction unit 16 performs a relay section specifying process (step S5-3).

  In step S5-3, first, the local reporter relay section extraction unit 16 goes back to the m sentence (for example, m = 50) by one sentence before the sentence S (i) that is the “relay closing comment”. Search for “Relaying Swing Comment”. That is, the local reporter relay section extraction unit 16 refers to the sentence analysis data, and increases k by 1 from 1 to m, while Kind2 (S (i−k)) is a value indicating “relay swing comment”. Search for S (ik). The sentence S (ik) of the “relaying comment” is found, and the municipality name specified for the sentence S (ik) is different from the municipality name specified for the sentence S (ik). If there is not, the local reporter relay section extraction unit 16 determines the sentence S (i-k) of “relay swing comment” to the sentence S (i) of “relay closing comment” as the “local reporter relay section”. The municipality name specified for the sentence S (i) and the municipality name specified for the sentence S (ik) are respectively the record of the sentence number i of the sentence analysis data and the sentence number (ik). Read from the record. In addition, the local reporter relay section extraction unit 16 determines that the municipality name is not specified for at least one of the sentence S (i) or the sentence S (ik), or “Ofunato-cho, Ofunato-shi, Iwate” and “ Even if there is an inclusive relationship in the city name, such as “Iwate Prefecture”, it is determined that there is no difference in the city name. The same applies to the case where it is determined in the following processing that there is no difference in city name or place name. The local reporter relay section extraction unit 16 determines that the relay location is the location of the municipality specified for the sentence S (i) or the sentence S (ik). The local reporter relay section extraction unit 16 newly generates relay section data and registers it in the storage unit 10 when the relay section data has not been generated yet. The local reporter relay section extraction unit 16 adds a record to the relay section data, forms the relay indicating “local reporter relay”, the start position indicating the sentence number (ik), and the end position indicating the sentence number k. And the location of the relay.

  If the “relay swing comment” sentence is not found even after going back to m sentences, the local reporter relay section extracting unit 16 reads m sentences one sentence before the sentence S (i) that is “relay closing comment” one sentence at a time. Go back to the sentence (for example, m = 50) and search for the sentence “Comment with relay”. That is, the local reporter relay section extraction unit 16 refers to the sentence analysis data, and increases k by 1 from 1 to m, while Kind2 (S (i−k)) is a value indicating “relayed comment”. Search for S (ik). When the sentence S (i-k) of “Comment with relay” is found, the local reporter relay section extraction unit 16 converts the sentence S (i-k) of “Comment with relay” to the sentence S ( It is determined that “i) is the“ local reporter relay section ”. Further, the local reporter relay section extraction unit 16 determines that the relay location is the location of the city name specified for the sentence S (i) or the sentence S (ik). The local reporter relay section extraction unit 16 adds a record to the relay section data, forms the relay indicating “local reporter relay”, the start position indicating the sentence number (ik), and the end position indicating the sentence number k. And the location of the relay.

  If neither the “relaying comment” sentence nor the “relayed comment” sentence is found even after going back to m sentences, the local reporter relay section extraction unit 16 refers to the news text and precedes the sentence S (i). The sentence is traced back to m sentences one sentence at a time (for example, m = 50), and a sentence S (ik) having a sentence end expression “is a place name” is searched (k is an integer from 1 to m). When a sentence S (ik) with the sentence end expression “is a place name” is found, and the place name included in the sentence is not different from the city name specified for the sentence S (i), The local reporter relay section extraction unit 16 determines that the sentence from the sentence S (ik) to the sentence S (i) of “relay closing comment” is the local reporter relay section. The local reporter relay section extraction unit 16 specifies the location of the relay from the location of the municipality specified for the sentence S (i) or the place name included in the sentence S (ik) (or from the place name) It is determined that this is the location of the city. The local reporter relay section extraction unit 16 adds a record to the relay section data, forms the relay indicating “local reporter relay”, the start position indicating the sentence number (ik), and the end position indicating the sentence number k. And the location of the relay.

  After the process of step S5-3, or when it is determined in step S5-2 that the sentence S (i) is not a “relay closing comment” (step S5-2: NO), the local reporter relay section extraction unit 16 Then, the process of step S5-4 is performed. That is, the local reporter relay section extraction unit 16 determines whether or not all sentences included in the news text have been processed by determining whether i is greater than 1 (step S5-4). If the local reporter relay section extraction unit 16 determines that i is greater than 1 and there is an unprocessed sentence (step S5-4: YES), 1 is subtracted from i and the sentence to be processed is moved forward. Shift by one (step S5-5) and repeat the processing from step S5-2. If the local reporter relay section extraction unit 16 determines that i has reached 1 and all the sentences indicated by the news text have been processed (step S5-4: NO), the process proceeds to step S5-6. .

  The local reporter relay section extraction unit 16 determines whether or not the i-th sentence S (i) to be processed is a “relay swing comment”, Kind2 (S (i) set in the sentence analysis data ) Is a value indicating “relaying comment” (step S5-6). If the local reporter relay section extraction unit 16 determines that the sentence S (i) is a “relay swing comment” (step S5-6: YES), the local reporter relay section extraction unit 16 performs a relay section specifying process (step S5-7).

  In step S5-7, the local reporter relay section extraction unit 16 advances the sentence after the sentence S (i), which is the “relaying comment”, to one sentence at a time (for example, m = 50). A comment sentence S (i + k) pointing to a “special comment” or a break in a news program is searched (k is an integer of 1 to m). Whether or not it is a “location specific comment” is determined by whether or not Kind1 (S (i + k)) set in the sentence analysis data is a value indicating “location specific comment”. In addition, comments that indicate breaks in news programs are, for example, comments such as “It was a relay” or “I will continue to tell you”. When a comment S (i + k) that indicates a “location specific comment” or a break in a news program is found, the local reporter relay section extractor 16 reads the sentence S (i) from the “relay swing comment” from the sentence S (i). Up to the sentence S (i + k−1) immediately before S (i + k) is determined as the local reporter relay section. The local reporter relay section extraction unit 16 identifies the location of the municipality specified for the sentence S (i) as the location of the relay, or the municipality specified for the sentence S (i + k−1) of the “location specific comment”. Judged to be the location of the town / village name. The local reporter relay section extraction unit 16 adds a record to the relay section data, forms the relay indicating “local reporter relay”, the start position indicating the sentence number i, and the end position indicating the sentence number (i + k−1). And the location of the relay.

  After the process of step S5-7, or when it is determined in step S5-6 that the sentence S (i) is not a “relaying comment” (step S5-6: NO), the local reporter relay section extracting unit 16 Step S5-8 is performed. That is, the local reporter relay section extraction unit 16 determines whether all sentences included in the news text have been processed by determining whether i is smaller than n (step S5-8). When the local reporter relay section extraction unit 16 determines that i is smaller than n and there is an unprocessed sentence (step S5-8: YES), 1 is added to i and the sentence to be processed is moved backward. Shift by one (step S5-9) and repeat the processing from step S5-6. When the local reporter relay section extraction unit 16 determines that i has reached n and all the sentences indicated by the news text have been processed (step S5-8: NO), step S1-4 in FIG. End the process.

  As described above, the local reporter relay section extraction unit 16 first extracts a local reporter relay section using a relay closing comment with high detection accuracy, and for the local reporter relay section where the relay closing comment is not detected, In addition, it is extracted by using a relay comment with high detection accuracy.

[Step S1-5: Reporter Absent Relay Section Extraction Process]
In step S1-5 of FIG. 5, the reporter absent relay section extraction unit 17 performs the processing result of the location specifying process (step S1-1), the processing result of the first sentence classification process (step S1-2), and the local reporter. Based on the processing result of the relay section extraction process (step S1-4), the reporter absent relay section and the relay location are extracted. This detailed processing will be described with reference to FIG.

FIG. 10 is a flowchart showing the processing procedure of the reporter absent relay section extraction process by the reporter absent relay section extraction unit 17, and shows the detailed processing procedure of step S1-5 in FIG.
First, the reporter absent relay section extraction unit 17 substitutes 1 for i in order to set a sentence to be processed as a sentence head (step S6-1). Next, the reporter absent relay section extraction unit 17 determines whether or not the local reporter relay section includes the i-th sentence S (i) to be processed in the relay section data as “relay reporter local reporter”. It is determined based on whether or not the sentence number i is included from the start position to the end position set in association with “relay” (step S6-2). When the reporter absent relay section extraction unit 17 determines that the sentence S (i) is not included in the local reporter relay section (step S6-2: NO), the sentence S (i) is “location specific comment”. Whether or not Kind1 (S (i)) set in the sentence analysis data is a value indicating “location specifying comment” (step S6-3). When the reporter absent relay section extraction unit 17 determines that the sentence S (i) is a “location specifying comment” (step S6-3: YES), the reporter absent relay section extraction unit 17 performs a relay section specifying process (step S6-4).

  In step S6-4, first, the reporter absent relay section extraction unit 17 advances the sentence after the sentence S (i) which is the “location specifying comment” to m sentences one sentence at a time (for example, m = 50). Search for the sentence “Location Specific Comment”. That is, the reporter absent relay section extraction unit 17 refers to the sentence analysis data, and increases the k from 1 to m by 1 while the Kind 1 (S (i + k)) is a value indicating the “location specifying comment”. Search for i + k). When the sentence S (i + k) of the “location specific comment” is found and there is no difference between the municipality name specified for the sentence S (i) and the municipality name specified for the sentence S (i + k), The reporter absent relay section extraction unit 17 determines that the sentence “location specific comment” from the sentence S (i) to the sentence S (i + k) is a “reporter absent relay section”. The municipality name specified for sentence S (i) and the municipality name specified for sentence S (i + k) are read from the sentence number i record and sentence number (i + k) record of sentence analysis data, respectively. It is. The reporter absent relay section extraction unit 17 determines that the relay location is the location of the city name specified for the sentence S (i) or the sentence S (i + k). The reporter absent relay section extraction unit 17 adds a record to the relay section data, forms the relay indicating “reporter absent relay”, the start position indicating the sentence number (i), and the end position indicating the sentence number (i + k). And the location of the relay.

  If a sentence “location specific comment” is not found even after searching up to m sentences, the reporter absent relay section extraction unit 17 selects one sentence after the sentence S (i) that is “location specific comment”. Step up to m sentences (for example, m = 50), and search for a sentence of “video comment”. In other words, the reporter absent relay section extracting unit 17 refers to the sentence analysis data, and increases k by 1 from 1 to m, while Kind1 (S (i + k)) is a value indicating “video comment”. ). If the sentence “S (i + k)” of the “video comment” is found and there is no difference between the municipality name specified for the sentence S (i) and the municipality name specified for the sentence S (i + k) The absent relay section extraction unit 17 determines that the sentence “location specific comment” from the sentence S (i) to the sentence S (i + k) is a “reporter absent relay section”. Further, the reporter absent relay section extraction unit 17 determines that the relay location is the location of the city name specified for the sentence S (i) or the sentence S (i + k). The reporter absent relay section extraction unit 17 adds a record to the relay section data, forms a relay indicating “reporter absent relay”, a start position indicating a sentence number i, an end position indicating a sentence number (i + k), Write the location of the relay.

  When it is determined in step S6-2 that the sentence S (i) is included in the local reporter relay section (step S6-2: YES), in step S6-3, the sentence S (i) is not a “location specific comment”. When it determines (step S6-3: NO), or after the process of step S6-4, the reporter absent relay area extraction part 17 performs the process of step S6-5. That is, the reporter absent relay section extraction unit 17 determines whether or not all sentences included in the news text have been processed by determining whether i is smaller than n (step S6-5). When the reporter absent relay section extraction unit 17 determines that i is smaller than n and there is an unprocessed sentence (step S6-5: YES), 1 is added to i and the sentence to be processed is moved backward. Shift by one (step S6-6) and repeat the processing from step S6-2. When the reporter absent relay section extracting unit 17 determines that i has reached n and all the sentences indicated by the news text have been processed (step S6-5: NO), the process proceeds to step S6-7. .

  The reporter absent relay section extraction unit 17 substitutes 1 for i in order to set the sentence to be processed as a sentence head (step S6-7). Next, the reporter absent relay section extraction unit 17 determines whether or not the i-th sentence S (i) to be processed is included in the local reporter relay section as in step S6-2 ( Step S6-8). When the reporter absent relay section extraction unit 17 determines that the sentence S (i) is not included in the local reporter relay section (step S6-8: NO), is the sentence S (i) a “video comment”? Whether or not Kind1 (S (i)) set in the sentence analysis data is a value indicating “video comment” is determined (step S6-9). When the reporter absent relay section extraction unit 17 determines that the sentence S (i) is a “video comment” (step S6-9: YES), the reporter absent relay section extraction unit 17 performs a relay section specifying process (step S6-10).

  In step S6-10, the reporter absent relay section extraction unit 17 advances the sentence after the sentence S (i), which is “video comment”, up to m sentences one sentence at a time (for example, m = 10). "Is searched. In other words, the reporter absent relay section extracting unit 17 refers to the sentence analysis data, and increases k by 1 from 1 to m, while Kind1 (S (i + k)) is a value indicating “video comment”. ). If the sentence “S (i + k)” of the “video comment” is found and there is no difference between the municipality name identified from the sentence S (i) and the municipality name identified from the sentence S (i + k) The absent relay section extraction unit 17 determines that the sentence “(video comment)” from the sentence S (i) to the sentence S (i + k) is a “reporter absent relay section”. The reporter absent relay section extraction unit 17 determines that the relay location is the location of the municipality specified from the sentence S (i) or the sentence S (i + k). The reporter absent relay section extraction unit 17 adds a record to the relay section data, forms a relay indicating “reporter absent relay”, a start position indicating a sentence number i, an end position indicating a sentence number (i + k), Write the location of the relay.

  When it is determined in S6-8 that the sentence S (i) is included in the local reporter relay section (step S6-8: YES), it is determined in step S6-9 that the sentence S (i) is not a “video comment”. In the case (step S6-9: NO) or after the process of step S6-10, the reporter absent relay section extracting unit 17 performs the process of step S6-11. That is, the reporter absent relay section extraction unit 17 determines whether all sentences included in the news text have been processed by determining whether i is smaller than n (step S6-11). When the reporter absent relay section extraction unit 17 determines that i is smaller than n and there is an unprocessed sentence (step S6-11: YES), 1 is added to i and the sentence to be processed is moved backward. Shift by one (step S6-12) and repeat the processing from step S6-8. When the reporter absent relay section extraction unit 17 determines that i has reached n and all the sentences indicated by the news text have been processed (step S6-11: NO), step S1-5 in FIG. End the process.

  As described above, the reporter absent relay section extraction unit 17 first extracts a reporter absent relay section using a location specific comment with high detection accuracy, and a video for a reporter absent relay section in which no location specific comment is detected. Extract using comments. Moreover, when the section extracted as the reporter absent relay section extracted by the reporter absent relay section extraction unit 17 has already been extracted as the local reporter relay section, the local reporter relay section is prioritized.

[Step S1-6: Aerial Shooting Section Extraction Processing]
In step S1-6 of FIG. 5, the aerial shooting relay section extraction unit 18 extracts the relay section extracted as a result of the local reporter relay section extraction process (step S1-4) and the reporter absent relay section extraction process (step S1-5). To extract the aerial shooting section. Therefore, words and phrases used for comments specific to aerial photography, such as “from the sky”, “video from a helicopter”, and “photographed by a helicopter” are registered in the storage unit 10 in advance. The aerial shooting section extraction unit 18 includes news text corresponding to the local reporter relay section extracted in step S1-4, news text corresponding to the reporter absent relay section extracted in step S1-5, and aerial shooting. Compare with specific comment words. Specifically, the aerial shooting relay section extraction unit 18 reads the start position and end position of the relay section for each record of the relay section data stored in the storage unit 10 until the read start position and end position. It is determined whether or not the sentence indicated by the news text corresponding to this section includes a comment phrase unique to aerial photography. When the aerial shooting section extraction unit 18 determines that a commentary phrase unique to aerial shooting is included, the relay form of the record is changed from “local reporter relay” or “reporter absent relay” to “aerial shooting”. Modify to “Relay section”.

[Step S1-7: Noise removal processing]
In step S1-7, the noise removing unit 19 performs local reporter relay section extraction processing (step S1-4), reporter absent relay section extraction processing (step S1-5), and aerial shooting relay section extraction processing (step S1-6). The noise is excluded from the relay section extracted as a result of (2) based on the length of time of the relay section. Specifically, the noise removal unit 19 starts the relay section (local reporter relay section, reporter absent relay section, or aerial shooting relay section) from each record of the relay section data stored in the storage unit 10. And the end position is read. The noise removal unit 19 acquires synchronization data corresponding to the sentence indicated by the start position and synchronization data corresponding to the sentence at the end position from the synchronization data added to the news text stored in the storage unit 10. The noise removal unit 19 subtracts the reproduction time indicated by the synchronization data corresponding to the sentence at the start position from the reproduction time indicated by the synchronization data corresponding to the sentence at the end position, and calculates the length of time of each relay section. A relay section in which the calculated time length is equal to or less than a certain value (for example, 10 seconds or less) is determined as noise. The noise removing unit 19 deletes the records of the local reporter relay section, the reporter absent relay section, and the aerial photography relay section determined as noise from the relay section data.

  In addition, the noise removal part 19 may perform the process which makes a continuous relay area the one relay area, before determining noise. For example, when records R1 and R2 having different relay section data have the same form of relay and the end position of record R1 and the start position of record R2 are continuous, the end position of record R1 is set to record R2. The record R2 is deleted at the end position.

  When the video data is a news program including a plurality of topics, the relay section extraction device 1 may extract a relay section in the section of video data of the same topic. In this case, the relay section extraction device 1 extracts the anchor shot from the video data by the conventional anchor shot detection method, and divides the video data so that the portion from which the anchor shot is extracted becomes the head of each topic.

  According to the above-described embodiment, the relay section extraction device extracts a phrase related to a place such as a place name or a landmark name from each sentence indicated by text data attached to a video of a news program, and specifies the place. Further, the relay section extraction device determines whether each sentence indicated by the text data is a comment specifying the location of the relay, or a comment explaining the video of the relay, and a comment with relay, a relay swing comment, or a relay closing comment. Is determined. The relay section extractor is based on the location specified from each sentence and the result of sentence classification processing for each sentence. Sections of scenes, such as relays from aerial photography from helicopters, etc., and relay locations are extracted. As a result, it is possible to acquire information as to which news program was reported from which location and in what form. By mapping the obtained results on a map, it is possible to visually grasp where and how long these relays have been broadcast, leading to information organization in the news media.

  Note that the relay section extraction device 1 described above has a computer system therein. The operation process of each unit of the relay section extraction device 1 is stored in a computer-readable recording medium in the form of a program, and the above-described processing is performed by the computer system reading and executing this program. The computer system here includes a CPU, various memories, an OS, and hardware such as peripheral devices.

Further, the “computer system” includes a homepage providing environment (or display environment) if a WWW system is used.
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. Furthermore, 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. In this case, a volatile memory in a computer system serving as a server or a client in that case, and a program that holds a program for a certain period of time are also included. 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.

DESCRIPTION OF SYMBOLS 1 Relay area extraction apparatus 10 Memory | storage part 11 Location identification part 12 Sentence classification part 13 First sentence classification part 14 Second sentence classification part 15 Relay area extraction part 16 Local reporter relay area extraction part 17 Reporter absence relay area extraction part 18 Aerial photography Relay section extraction unit 19 Noise removal unit

Claims (7)

A place specifying unit that extracts a phrase related to a place from each sentence indicated by text data associated with video data of a program, and specifies a place from the extracted phrase;
A sentence classification unit for determining whether each sentence indicated by the text data is a comment using an expression related to relay;
The type of the comment for each of the two different sentences determined to be comments using the relay-related expression by the sentence classification unit and the location specified by the place specifying unit for each of the two different sentences is different. Based on the heel, a relay section extraction unit that extracts the section of the video data corresponding to the two different sentences as a relay section;
A relay section extraction device comprising: The sentence classification unit determines whether each sentence indicated by the text data is a relay entry comment, a relay swing comment, and a relay closing comment.
The relay section extraction unit is a location identified by the location identification unit for a sentence determined to be a relay closing comment, appears before the sentence in the text data, and is a relay swing comment It is determined that there is no difference between the determined location of the other sentence and the location specified by the location specifying unit, or that the comment is a relayed comment before the sentence determined to be a relay closing comment in the text data. When the recorded sentence appears, the video data section corresponding to the sentence determined to be the relay closing comment from the other sentence determined to be the relay comment or the relayed comment Extract as a relay section by a local reporter,
The relay section extracting device according to claim 1, wherein The sentence classification unit further determines whether or not each sentence indicated by the text data is a comment that tells the location of the relay,
The relay section extraction unit determines that the sentence is determined to be a relay comment, the place specified by the place specifying unit, and a comment that appears after the sentence in the text data and conveys the place If there is no difference between the place specified by the place specifying unit and the other sentence that has been determined, it is determined that the comment is a comment that conveys the place from the sentence that is determined to be a relay comment. The corresponding video data section is extracted as a relay section by a local reporter.
The relay section extraction device according to claim 2, wherein The sentence classification unit determines whether or not each sentence indicated by the text data is a comment that tells a relay location, and whether or not it is a comment explaining a relay video,
The relay section extraction unit identifies a location identified by the location identification unit for a sentence determined to be a comment that conveys a relay location, and a comment that appears after the sentence in the text data and conveys the relay location Alternatively, if there is no difference between the place specified by the place specifying unit for other sentences determined to be comments explaining the video of the relay, the sentence determined to be a comment that conveys the place and the location of the relay The reporter extracts the section of the video data corresponding to the other sentence determined to be a comment explaining the video or the video of the relay as a relay section that is absent from the site,
The relay section extraction device according to any one of claims 1 to 3, wherein When there is no difference in the location specified by the location specifying unit for each of two different sentences determined to be comments explaining the video of the relay, the relay section extraction unit corresponds to the two different sentences. The section of the video data is extracted as a relay section where the reporter is not present locally.
The relay section extracting device according to claim 4, wherein Aerial relay for determining that the relay section is a relay section by aerial photography when the sentence in the text data corresponding to the relay section extracted by the relay section extraction unit includes a word representing aerial photography Further comprising a section extraction unit,
The relay section extraction device according to any one of claims 1 to 5, wherein A computer used as a relay section extraction device,
A location specifying unit for extracting a phrase related to a location from each sentence indicated by text data associated with video data of a program, and specifying the location from the extracted phrase;
A sentence classification unit for determining whether or not each sentence indicated by the text data is a comment using an expression related to relay;
The type of the comment for each of the two different sentences determined to be comments using the relay-related expression by the sentence classification unit and the location specified by the place specifying unit for each of the two different sentences is different. Based on the heel, a relay section extraction unit that extracts the section of the video data corresponding to the two different sentences as a relay section,
Program to function as.

Images