Method and apparatus for multiple tiered matching of natural language queries to positions in a text corpus

• US 6,560,590 B1
• Filed: 02/14/2000
• Issued: 05/06/2003
• Est. Priority Date: 02/14/2000
• Status: Expired due to Term

• Current Assignee: Verint Americas Incorporated (Verint Systems Incorporated)
• Original Assignee: KANA Software Incorporated (Verint Systems Incorporated)
• Inventors: OʼNeill, Brian, Shwe, Michael
• Primary Examiner(s): Starks, Jr., Wilbert L.
• Assistant Examiner(s): Hirl, Joseph P.
• Application Number:
  US09/503,983
• Time in Patent Office: 1,177 Days
• Field of Search: 706/55, 706/45, 706/46, 707/2
• US Class Current: 706/55
• CPC Class Codes: G06F 16/3344 using natural language analysis
• CPC Class Codes: G06F 16/951 Indexing; Web crawling techniques
• CPC Class Codes: Y10S 707/99932 Access augmentation or optimizing

Abstract

A multi-tiered system for responding to natural language queries is using disclosed. The query resolution system returns zero or more links to content that is relevant to the users query. The present invention for query resolution combines two or more types of natural language query resolution methods, where the knowledge base for each of the methods comes from a single knowledge specification. The various different natural language query resolution methods differ fundamentally in how they match the user query to the web site content. The results of the resolution methods are ranked and all, some, or none of the results of each system may be displayed.

Claims

1. A computer implemented method of processing natural language queries, said method comprising:

invoking a metareasoning module to obtain a set of expected result metrics for a plurality of different natural language query resolution methods, each of said different natural language query resolution methods capable of returning a final natural language query result;

selecting a natural language query resolution method from said plurality of different natural language query resolution methods dependent on said set of expected result metrics for said plurality of different natural language query resolution methods; and

invoking said selected natural language query resolution method to obtain actual natural language query results.

2. The method of processing natural language queries as claimed in claim 1 further comprising:

returning at least one tag or pointer to a section of an information corpus from said actual natural language query results.

3. The method of processing natural language queries as claimed in claim 1 wherein selecting a natural language query resolution method is further dependent upon additional factors.

4. The method of processing natural language queries as claimed in claim 3 wherein one of said additional factors comprises a current load.

5. The method of processing natural language queries as claimed in claim 1 wherein said method repeats selecting said natural language query resolution method and invoking said selected natural language query resolution method until a set of desired natural language query results are fulfilled by said actual natural language query results.

6. The method of processing natural language queries as claimed in claim 1, said method further comprising:

analyzing said actual natural language query results;

selecting a next natural language query resolution method dependent on said actual natural language query results from at least one previous natural language query resolution method.

8. The method of processing natural language queries as claimed in claim 6 wherein selecting said next natural language query resolution method is further dependent upon additional factors.

9. The method of processing natural language queries as claimed in claim 8 wherein one of said additional factors comprises a current load.

10. The method of processing natural language queries as claimed in claim 1, said method further comprising:

selecting a next natural language query resolution method from said plurality of different natural language query resolution methods; and

invoking said next natural language query resolution method wherein said next natural language query resolution method is provided with a set of partial results from said previous selected natural language method.

11. The method of processing natural language queries as claimed in claim 1 wherein said method further comprises:

adjusting a Bayesian network knowledge base used by a natural language query resolution method after a natural language query has been processed.

7. The method of processing natural language queries as claimed in 6 wherein said method repeats selecting said natural language query resolution method dependent on said set of expected result metrics and invoking said selected natural language query resolution method until a set of desired natural language query results are fulfilled by said actual natural language query results.

12. A computer implemented method of processing a natural language queries, said method comprising:

invoking a first natural language query resolution method from a plurality of different natural language query resolution methods to obtain partial results, each of said different natural language query resolution methods in said plurality of different natural language query resolution methods capable of returning a final natural language query result; and

invoking a next natural language query resolution method from said plurality of different natural language query resolution methods using said partial results from said first natural language query resolution method.

13. The method of processing natural language queries as claimed in claim 12 further comprising:

invoking a metareasoning module to obtain a set of expected result metrics for a plurality of natural language query resolution methods; and

selecting said first natural language query resolution method from a plurality of natural language query resolution methods dependent on said set of expected result metrics.

14. The method as claimed in claim 13 wherein selecting said natural language query resolution method is further dependent upon additional factors.

15. The method a s claimed in claim 12 further comprising:

analyzing said partial results;

selecting said next natural language query resolution method dependent on said partial results from said first natural language query resolution method.

16. A computer implemented natural language processing system for processing a natural language query, said natural language processing system comprising:

a single main knowledge representation;

a first natural language query resolution method for analyzing said natural language query, said first natural language query resolution method capable of returning a first final natural language query result;

a first knowledge base for said first natural language query resolution method, said first knowledge base derived from said single main knowledge representation;

a second natural language query resolution method for analyzing said natural language query, said second natural language query resolution method using a second knowledge base;

a second knowledge base for said second natural language query resolution method, said second knowledge base derived from said single main knowledge representation, said second natural language query resolution method capable of returning a second final natural language query result; and

a metareasoner for invoking said first or second natural language query resolution methods based upon expected result metrics.

17. The natural language processing system as claimed in claim 16 wherein said single main knowledge base comprises a set of patterns and an expansion of said set of patterns.

18. The natural language processing system as claimed in claim 17 further comprising:

a derivation system, said derivation system generating an uncertainty based reasoning system based upon said set of patterns and said expansion of said set of patterns.

19. The natural language processing system as claimed in claim 16 wherein said single main knowledge base comprises a grammar.

20. The natural language processing system as claimed in claim 19 wherein said first natural language query resolution method comprises a grammar based method and said first knowledge base comprises a parse tree derived from said regular grammar.

21. The natural language processing system as claimed in claim 19 wherein said first natural language query resolution method comprises a Bayesian network based method and said first knowledge base comprises a Bayesian network derived from said regular grammar.

22. The system as claimed in claim 16 further comprising:

a refinement system, said refinement system improving said single main knowledge representation based upon results from a final result from a natural language query.

24. The method of processing natural language queries as claimed in claim 22 wherein said expected metrics comprise expected recall value.

23. A computer implemented method of processing natural language queries, said method comprising:

consulting a metareasoning knowledge base to compare expected metrics for a plurality of natural language query resolution methods, each of said different natural language query resolution methods in said plurality of different natural language query resolution methods capable of returning a final natural language query result;

selecting a natural language query resolution method from said plurality of natural language query resolution methods dependent on information from said metareasoning knowledge base; and

invoking said selected natural language query resolution method to obtain actual results.

25. The method of processing natural language queries as claimed in claim 23 wherein said expected metrics comprise expected precision value.

26. The method of processing natural language queries as claimed in claim 25 wherein said method successively invokes natural language query resolution methods in an order of decreasing precision values.