Topic transition analysis system, method, and program

• US 8,670,978 B2
• Filed: 12/14/2009
• Issued: 03/11/2014
• Est. Priority Date: 12/15/2008
• Status: Active Grant

• Current Assignee: NEC Corporation
• Original Assignee: NEC Corporation
• Inventors: Nagatomo, Kentaro
• Primary Examiner(s): ALBERTALLI, BRIAN LOUIS
• Application Number:
  US13/139,550
• Publication Number:
  US 20110246183A1
• Time in Patent Office: 1,548 Days
• Field of Search: None
• US Class Current: 704/9
• CPC Class Codes: G06F 16/36 Creation of semantic tools, e.g. ontology or thesauri

Abstract

The present invention provides a topic transition analysis system that determines a position on a primary media stream leading to a certain statement made in a language communication carried out in a secondary channel associated with the primary media stream. The topic transition analysis system includes a statement trigger string determination unit receiving a primary media stream and one or a plurality of language communication streams (hereinafter, language streams) executed in parallel with the media stream and determining whether or not a certain statement on the one or plurality of language streams has been made newly in response to contents of the media stream.

Claims

1. A topic transition analysis system comprising:

a statement trigger string determination unit receiving a primary media stream and one or a plurality of language communication streams (hereinafter, language streams) executed in parallel with the media stream and determining whether or not a certain statement on the one or plurality of language streams has been made newly in response to contents of the media stream,

wherein the statement trigger string determination unit executes the determination based on a higher one of a similarity between the statement and media stream contents at and around the time when the statement is made and a similarity between the statement and language stream contents prior to the time when the statement is made.

2. A media stream indexing system comprising:

the topic transition analysis system according to claim 1; and

a unit estimating a point or section on the media stream triggering an arbitrary statement on the language stream.

3. A language stream search system comprising:

the topic transition analysis system according to claim 1; and

a unit estimating a statement string on the language stream originating from an arbitrary point or section on the media stream.

4. The topic transition analysis system according to claim 1 comprising:

at least one of topic similarity calculation unit calculating a topic similarity between the statement and a media stream at and around the time when the statement is made and a topic similarity between the statement and a group of statements made immediately before the statement,

wherein the statement trigger string determination unit determines a string having a higher topic similarity to be a string that has triggered the statement.

5. The topic transition analysis system according to claim 4,

wherein said at least one of the topic similarity calculation unit receives the statement and one or a plurality of statement groups which are similarity calculation targets, calculates document vectors with respect to the statement and the statement groups by adding weight to each term included in the statement and the statement groups based on a given algorithm, calculating distances among these document vectors, and calculating topic similarities based on the distances.

6. The topic transition analysis system according to claim 5, further comprising:

a speech recognition unit generating a text string from a speech signal included in the primary media stream,

wherein said at least one of the topic similarity calculation unit calculates a topic similarity between the statement and a text string generated by the speech recognition unit from the entire section or an arbitrary section of the media stream which is a similarity calculation target.

7. The topic transition analysis system according to claim 4, further comprising:

a speech recognition unit generating a text string from a speech signal included in the primary media stream,

wherein said at least one of the topic similarity calculation unit calculates a topic similarity between the statement and a text string generated by the speech recognition unit from the entire section or an arbitrary section of the media stream which is a similarity calculation target.

8. A topic transition analysis system comprising:

a statement trigger string determination unit receiving a primary media stream and one or a plurality of language communication streams (hereinafter, language streams) executed in parallel with the media stream and determining whether or not a certain statement on the one or plurality of language streams has been made newly in response to contents of the media stream; and

one or a plurality of topic similarity calculation unit calculating a topic similarity between the statement and a media stream at and around the time when the statement is made and a topic similarity between the statement and a group of statements made immediately before the statement,

wherein the statement trigger string determination unit determines a string having a higher topic similarity to be a string that has triggered the statement.

9. The topic transition analysis system according to claim 8,

wherein at least one of the topic similarity calculation unit receives the statement and one or a plurality of statement groups which are similarity calculation targets, calculates document vectors with respect to the statement and the statement groups by adding weight to each term included in the statement and the statement groups based on a given algorithm, calculating distances among these document vectors, and calculating topic similarities based on the distances.

10. The topic transition analysis system according to claim 9, further comprising:

a speech recognition unit generating a text string from a speech signal included in the primary media stream,

wherein said at least one of the topic similarity calculation unit calculates a topic similarity between the statement and a text string generated by the speech recognition unit from the entire section or an arbitrary section of the media stream which is a similarity calculation target.

11. The topic transition analysis system according to claim 8, further comprising:

a speech recognition unit generating a text string from a speech signal included in the primary media stream,

wherein at least one of the topic similarity calculation unit calculates a topic similarity between the statement and a text string generated by the speech recognition unit from the entire section or an arbitrary section of the media stream which is a similarity calculation target.

12. The topic transition analysis system according to claim 11 outputting speech recognition results of a string that is determined to have a higher topic similarity by the statement trigger string determination unit.

13. The topic transition analysis system according to claim 8, further comprising:

a topic group storage unit classifying a plurality of statements appearing on the language stream into topic groups according to topic similarity and storing the plurality of statements according to the topic groups,

wherein the topic similarity calculation unit calculates a topic similarity between the statement and each of the groups accumulated in the topic group storage unit, and

wherein the statement trigger string determination unit determines a string having a higher topic similarity to be a string that has triggered the statement.

14. The topic transition analysis system according to claim 8, further comprising:

a speech recognition unit generating a text string adapted for an arbitrary topic from a speech signal included in the language stream; and

a speech recognition result reliability estimation unit estimating validity of a text string generated by the speech recognition unit,

wherein, when the speech recognition unit generates a text string from a speech signal, which is a speech of the statement, of a language stream section, at least one of the topic similarity calculation unit executes a speech recognition process adapted for a topic estimated from the entire section or an arbitrary section of the media stream and one or a plurality of statement groups, which are similarity calculation targets, and calculates a topic similarity based on reliability indices obtained after the speech recognition result reliability estimation unit estimates validity on the text strings obtained by the speech recognition process.

15. The topic transition analysis system according to claim 8, further comprising:

a topic transition network generation unit generating, based on the topic similarity, a topic transition network in which arbitrary points or sections of the media stream and individual statement trigger strings are represented in the form of a tree-shaped network.

16. A topic transition analysis method comprising:

selecting, by a computer, an arbitrary statement from one or a plurality of language communication streams (hereinafter, language streams) executed in parallel with a media stream; and

determining, by the computer, whether or not the statement has been made newly in response to contents of the media stream,

wherein the determination is executed based on a higher one of a similarity between the statement and media stream contents at and around the time when the statement is made and a similarity between the statement and language stream contents prior to the time when the statement is made.

17. A non-transitory computer-readable data storage medium storing a program causing a computer configuring a topic transition analysis system to execute the processes of:

selecting an arbitrary statement from one or a plurality of language communication streams (hereinafter, language streams) executed in parallel with a media stream; and

determining whether or not the statement has been made newly in response to contents of the media stream,

wherein determining whether or not the statement has been made newly in response to contents of the media stream is based on a higher one of a similarity between the statement and media stream contents at and around the time when the statement is made and a similarity between the statement and language stream contents prior to the time when the statement is made.