System and method of automating a spoken dialogue service

• US 8,566,102 B1
• Filed: 11/06/2002
• Issued: 10/22/2013
• Est. Priority Date: 03/28/2002
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: AT&T Intellectual Property II LP (AT&T, Inc.)
• Inventors: Bangalore, Srinivas, Feng, Junlan, Rahim, Mazin G
• Primary Examiner(s): JACKSON, JAKIEDA R
• Application Number:
  US10/288,764
• Time in Patent Office: 4,003 Days
• Field of Search: 704/270.1
• US Class Current: 704/270.1
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 2015/228 of application context

Abstract

A system and method of generating and operating a spoken dialog service for a web-site are disclosed. The system parses web-site data and organizes the web-site data in a task knowledge data bank. The system receives text associated with a user query; processes the received text in a spoken language understanding (SLU) module, the SLU module using the web-site data from the task knowledge data bank; generates a ranked list of relevant responses to the user query; generates a hierarchical tree using the web-site data and the ranked list of relevant responses to the user query, generates a response to the user query using the hierarchical tree; and presents the response to the user.

Claims

1. A method comprising:

extracting, via a processor and independent of user input, first task data from a web-site, wherein the first task data is based on a structure of a web page in the web-site;

formatting the first task data into first formatted task data comprising a first topic section, a first hyperlink section, and a first content/answer section;

extracting, independent of the user input, second task data from the web-site, wherein the second task data is based on the structure of the web page in the web-site;

formatting the second task data into second formatted task data comprising a second topic section, a second hyperlink section, and a second content/answer section;

analyzing the first formatted task data and the second formatted task data, to yield an analysis;

when, based on the analysis, the first topic section is determined to be thematically coherent with the second topic section, merging the first formatted task data and the second formatted task data into third formatted task data;

receiving a spoken natural language user query;

organizing task data into a ranked hierarchical structure based on the spoken natural language user query, wherein the task data comprises one of (1) the first formatted task data and the second formatted task data, and (2) the third formatted task data;

generating a ranked list of relevant responses to the spoken natural language user query using the ranked hierarchical structure to perform vector space modeling; and

initiating a two-way, natural language spoken dialog to provide a response to the spoken natural language user query according to the ranked list.

2. The method of claim 1, wherein extracting of the first task data and extracting of the second task data further comprise extracting text from the web-site.

3. The method of claim 2, wherein the text is extracted in text segments.

4. The method of claim 3, wherein merging of the first formatted task data and the second formatted task data further comprises integrating thematically the first formatted task data and the second formatted task data.

5. The method of claim 1, wherein organizing the task data further comprises representing the task data using a plurality of parameters.

6. The method of claim 5, wherein the plurality of parameters comprises a structure code, a parent tag, a text code, a color code, a size, and a link code.

7. The method of claim 6, wherein extracting of the first task data and extracting of the second task data further comprise extracting text from the web-site.

8. The method of claim 7, wherein the text is extracted in text segments.

9. The method of claim 8, further comprising:

identifying themes with the text segments that have been extracted.

10. The method of claim 9, further comprising:

determining topic and content pairs associated with the text segments that have been extracted.

11. The method of claim 1, wherein extracting the first task data and the second task data further comprises extracting data relationship information.

12. The method of claim 1, further comprising:

deleting redundant information.

13. A method comprising:

detecting, via a processor, when new web-site data is added to a web-site, the new web-site data being based on a structure of a web page in the web-site;

extracting the new web-site data;

formatting the new web-site data into new formatted web-site data comprising a new topic section, a new hyperlink section, and a new content/answer section;

analyzing, via a processor, the new formatted web-site data and old formatted web-site data, to yield an analysis, the old formatted web-site data comprising an old topic section, an old hyperlink section, and an old content/answer section;

when, based on the analysis, the new topic section is determined to be thematically coherent with the old topic section, merging the new formatted web-site data and the old formatted web-site data into third formatted web-site data;

receiving a spoken natural language user query;

organizing web-site data into a ranked hierarchical structure based on the new web site data for the spoken natural language user query, wherein the web-site data comprises one of (1) the new formatted web-site data and the old formatted web-site data, and (2) the third formatted task data;

generating a ranked list of relevant responses to the spoken natural language user query, using the ranked hierarchical structure to perform vector space modeling; and

initiating a two-way, natural language spoken dialog to provide a response to the spoken natural language user query based on the ranked list.

14. A method comprising:

automatically detecting, via a processor, when additional web-site structure data is added to a web-site, wherein the additional web-site structure data is based on a structure of the web-site;

extracting the additional web-site structure data from the web-site;

formatting the additional web-site structure data into formatted additional data comprising an additional topic section, an additional hyperlink section, and an additional content/answer section;

analyzing, via a processor, the formatted additional data and formatted original data, to yield an analysis, wherein the formatted original data comprises an original topic section, an original hyperlink section, and an original content/answer section;

when, based on the analysis, the original topic section is determined to be thematically coherent with the additional topic section, merging the formatted original data and the formatted additional data into third formatted data;

receiving the spoken natural language user query;

organizing data into a ranked hierarchical structure based on the spoken natural language user query, wherein the data comprises one of (1) the formatted original data and the formatted original data, and (2) the third formatted data;

generating a ranked list of relevant responses to the spoken natural language user query using the data; and

synchronizing a two-way, natural language spoken dialog to provide a response to the spoken natural language user query according to the ranked list.

15. The method of claim 14, wherein extracting the additional web-site structure data comprises parsing the web-site data, determining relevance of text in the web-site data and determining a relationship of text in the web-site data.

16. A system comprising:

a processor; and

a computer-readable storage medium having instructions stored which, when executed on the processor, perform operations comprising;

extracting, via a processor and independent of user input, first task data from a web-site, wherein the first task data is based on a structure of a web page in the web-site;

formatting the first task data into first formatted task data comprising a first topic section, a first hyperlink section, and a first content/answer section;

extracting, independent of the user input, second task data from the web-site, wherein the second task data is based on the structure of the web page in the web-site;

formatting the second task data into second formatted task data comprising a second topic section, a second hyperlink section, and a second content/answer section;

analyzing the first formatted task data and the second formatted task data, to yield an analysis;

when, based on the analysis, the first topic section is determined to be thematically coherent with the second topic section, merging the first formatted task data and the second formatted task data into third formatted task data;

receiving a spoken natural language user query;

organizing task data into a ranked hierarchical structure based on the spoken natural language user query, wherein the task data comprises one of (1) the first formatted task data and the second formatted task data, and (2) the third formatted task data;

generating a ranked list of relevant responses to the spoken natural language user query-using the ranked hierarchical structure to perform vector space modeling;

and

initiating a two-way, natural language spoken dialog using the ranked list.

17. The system of claim 16, wherein organizing the task data into the ranked hierarchical structure further comprises:

extracting, prior to receiving the spoken natural language user query, text and text attributes from the task data.

18. The system of claim 17, wherein organizing the task data into the hierarchical structure further comprises:

identifying function hypertext markup language pages in the web-site.

19. The system of claim 18, wherein organizing the task data into the ranked hierarchical structure further comprises:

filtering content in the web-site.

20. The system of claim 19, wherein organizing the task data into the ranked hierarchical structure further comprises:

identifying a role of text in the web-site and a relationship between the text.

21. The system of claim 19, wherein filtering content in the web-site further comprises eliminating redundancy.

22. A computer-readable storage device having instructions which, when executed on a computing device, cause the computing device to perform operations comprising:

parsing web-site data independent of user input, the web-site data being based on a structure of a web-page in a web-site, yielding parsed web-site data;

formatting the parsed web-site data into formatted parsed web-site data comprising a parsed topic section, a parsed hyperlink section, and a parsed content/answer section such that a text segment from the web-page is extracted and represented using the task data;

analyzing the formatted parsed web-site data and formatted old web-site data, to yield an analysis, the formatted old web-site data comprising an old topic section, an old hyperlink section, and an old parsed content/answer section;

when, based on the analysis, the parsed topic section is determined to be thematically coherent with the old topic section, merging the formatted parsed web-site data and the formatted old web-site data into a third formatted web-site data;

receiving a spoken natural language user query;

organizing web-site data into a ranked hierarchical structure based on the spoken natural language user query, wherein the web-site data comprises one of (1) the formatted parsed web-site data and the formatted old web-site data, and (2) the third formatted web-site data;

generating a ranked list of relevant responses to the spoken natural language user query using the ranked hierarchical structure to perform vector space modeling;

and

using a spoken dialog model to generate a spoken, audible response to the spoken natural language user query based on the ranked list; and

initiating a two-way, natural language spoken dialog which delivers the spoken, audible response.

23. The computer-readable storage device of claim 22, wherein the two-way, natural language dialog is generated by summarizing a long text segment from the web-site while generating the spoken, audible response to the spoken natural language user query.

24. A system comprising:

a processor; and

a computer-readable storage medium having instructions stored which, when executed on the processor, cause the processor to perform operations comprising;

parsing web-site data, independent of user input, the web-site data being based on a structure of a web-site to yield parsed web-site data;

formatting the parsed web-site data, to yield formatted parsed web-site data comprising a parsed topic section, a parsed hyperlink section, and a parsed content/answer section;

analyzing the formatted parsed web-site data and formatted old web-site data, to yield an analysis, the formatted old web-site data comprising an old topic section, an old hyperlink section, and an old parsed content/answer section;

when, based on the analysis, the parsed topic section is determined to be thematically coherent with the old topic section, merging the formatted parsed web-site data and the formatted old web-site data into a third formatted web-site data;

receiving a spoken natural language user query;

organizing web-site data into a ranked hierarchical structure based on the spoken natural language user query, wherein the web-site data comprises one of (1) the formatted parsed web-site data and the formatted old web-site data, and (2) the third formatted web-site data;

generating a ranked list of relevant responses to the spoken natural language user query using the web-site data to perform vector space modeling; and

using the ranked list to generate a spoken, audible response to the spoken natural language user query from web-site content as part of a two-way natural language spoken dialog, wherein the two-way, natural language spoken dialog is configured for operating without translating web-pages of the web-site.

25. The system of claim 24, wherein the spoken, audible response to the spoken natural language user query from the web-site content comprises a portion of the web-site content associated with the spoken natural language user query.

26. The system of claim 24, further comprising summarizing long text segments to generate the spoken, audible response.

27. A method comprising:

parsing web-site data, via a processor and independent of user input, the web-site data being based on a structure of a web-site, to yield parsed web-site data;

formatting the parsed web-site data, to yield formatted parsed web-site data comprising a parsed topic section, a parsed hyperlink section, and a parsed content/answer section;

analyzing the formatted parsed web-site data and formatted old web-site data, to yield an analysis, the formatted old web-site data comprising an old topic section, an old hyperlink section, and an old parsed content/answer section;

when, based on the analysis, the parsed topic section is determined to be thematically coherent with the old topic section, merging the formatted parsed web-site data and the formatted old web-site data into third formatted web-site data;

receiving, via a processor, a spoken natural language user query;

organizing web-site data into a ranked hierarchical structure based on the spoken natural language user query, wherein the web-site data comprises one of (1) the formatted parsed web-site data and the formatted old web-site data, and (2) the third formatted web-site data;

generating a ranked list of relevant responses to the spoken natural language user query using data from the ranked hierarchical structure to perform vector space modeling;

generating a spoken, audible response to the spoken natural language user query using the ranked hierarchical structure; and

initiating a two-way natural language spoken dialog to provide a response to the spoken natural language user query based on the ranked list.

28. The method of claim 27, wherein the spoken, audible response is further generated using a portion of the parsed web-site data that relates to the spoken natural language user query.

29. The method of claim 28, wherein generating the spoken, audible response to the spoken natural language user query based on ranked hierarchical structure data further comprises generating a hierarchical tree.

30. The method of claim 29, wherein generating the spoken, audible response further comprises:

extracting text attributes;

identifying web-page functions;

filtering web-site content; and

identifying text relationships.

31. A method comprising:

receiving, via a processor, a spoken natural language user query;

receiving text associated with the spoken natural language user query;

receiving first web-site data based on a structure of a web-site;

formatting the first web-site data into first formatted web-site data comprising a first topic section, a first hyperlink section, and a first content/answer section;

receiving second web-site data based on a structure of the web-site;

formatting the second web-site data into second formatted web-site data comprising a second topic section, a second hyperlink section, and a second content/answer section;

analyzing the first formatted web-site data and the second formatted web-site data, to yield an analysis;

when, based on the analysis, the first topic section is determined to be thematically coherent with the second topic section, merging the first formatted web-site data and the second formatted web-site data into a third formatted web-site data;

organizing web-site data into a ranked hierarchical structure based on the spoken natural language user query and the text, wherein the web-site data comprises one of (1) the first formatted web-site data and the second formatted web-site data, and (2) the third formatted web-site data;

generating a ranked list of relevant response to the spoken natural language user query using the ranked hierarchical structure to perform vector space modeling;

generating a spoken, audible response to the spoken natural language user query using the ranked hierarchical tree via a two-way, natural language spoken dialog; and

presenting the spoken, audible response to the user.

32. The method claim 31, further comprising:

summarizing information from the web-site when generating the spoken, audible response.

33. The method of claim 31, wherein organizing of the web-site data into the ranked hierarchical structure further comprises receiving context information regarding a current dialog.