INTEGRATING SPECIALIZED KNOWLEDGE SOURCES INTO A GENERAL SEARCH SERVICE

• US 20120005148A1
• Filed: 06/30/2010
• Published: 01/05/2012
• Est. Priority Date: 06/30/2010
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Horvitz, Eric, Apacible, Johnson
• Granted Patent: US 8,732,222 B2
• Time in Patent Office: Days
• US Class Current: 706/50
• CPC Class Codes: G06F 16/337 Profile generation, learning or modification
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 5/043 Distributed expert systems; Blackboards

Abstract

Embodiments are described for a system for integrating expert sources of knowledge into a general search service. The method can include a general search engine interface to receive a search query with at least one search term from a user. An expert knowledge engine can enable retrieval of expert information and analyses. A predictive model module can be used to determine a probability that an expert knowledge engine may provide a valuable response to the query. The search query can be routed to the expert knowledge engine based on a cost-benefit analysis. A query reformulation module can reformulate a query for use by the expert knowledge engine. A display module can be employed to combine the expert search results with other search results.

Claims

1. A system for integrating a specialized knowledge source into a general search service, comprising:

a search engine interface to receive a search query having search terms from a user;

an expert knowledge engine configured to enable retrieval of expert search results;

a predictive model module configured to generate a predictive model that provides a measure of the likelihood that the expert knowledge engine can provide valuable information for the search query;

a query reformulation module to reformulate a query for use by the expert knowledge engine; and

a display module configured to display the expert search results to the user.

2. The system as in claim 1, further comprising a result prediction module configured to predict a confidence probability for the expert search results of the expert knowledge engine.

3. The system as in claim 2, wherein the expert search results are displayed to the user when the result prediction module predicts that the search results have a confidence probability that exceeds an operating threshold.

4. The system as in claim 1, wherein the predictive model module is configured to perform quantitative decision analysis using an expected value and an expected cost for sending a search query to the expert knowledge engine to form a net expected value, wherein the net expected value is used to determine when to send the search query to the expert knowledge engine.

5. The system as in claim 1, wherein the display module can display the expert search results in an expert search area along with general search results.

6. The system as in claim 1, wherein the expert knowledge engine includes expert information about a plurality of topics and further comprises a topic relevance module configured to predict a relevance of a topic in the expert knowledge engine once the search query has been determined to be relevant to the expert knowledge engine.

7. A system as in claim 1, wherein the case library includes search queries and query features that are tagged by success or failure of the expert knowledge engine to provide a valuable response, and the case library is used to construct a predictive model to determine a probability that the search query is relevant to the expert knowledge engine.

8. A system as in claim 1, wherein the query reformulation module reformulates the search query based on expert rules for the expert knowledge engine and topics.

9. A system as in claim 1, wherein the query reformulation module reformulates the search query so that a separate query re-write can be created for each of a plurality of topics in the expert knowledge engine.

10. The system as in claim 1, wherein the predictive model module is configured to determine a probability that the search terms are related to the expert knowledge engine and to send the search query the expert knowledge engine when the probability exceeds an operating threshold.

11. The system as in claim 10, wherein the operating threshold may vary based on operating threshold modifiers.

12. A method for integrating expert knowledge sources into general search services, comprising:

receiving a search query having search terms from a user;

identifying an expert knowledge engine configured to enable retrieval of expert data;

applying a predictive model to the search query to determine the relevance of an expert knowledge engine to the search query using a statistical model;

reformulating the search query to provide a format usable by the expert knowledge engine; and

obtaining an expert search result from the expert knowledge engine for the search query.

13. A method as in claim 12, further comprising:

testing the expert search result of the reformulated search query to determine a confidence probability; and

displaying the expert search result to the user based on the confidence probability exceeding a operating threshold.

14. The method as in claim 12, further comprising displaying the expert search results in an expert search area along with general search results.

15. The method as in claim 12, further comprising using a user action in response to the expert search result to further improve the accuracy of a selection of the expert knowledge engine for the search query.

16. The system as in claim 12, further comprising predicting a relevance of a topic in the expert knowledge engine once the search query has been determined to be relevant to a subject of the expert knowledge engine.

17. A system for integrating expert knowledge sources into a search service, comprising:

a search engine interface to receive a search query having search terms from a user;

an expert knowledge engine configured to enable retrieval of expert search results;

a case library having tagged search cases related to the search terms;

a predictive model module configured to determine a probability that the search terms are related to the expert knowledge engine and to send the search query the expert knowledge engine when the probability exceeds a operating threshold;

a query reformulation module to reformulate a query in a format usable by the expert knowledge engine;

a result prediction module configured to predict a confidence probability for the results of the expert knowledge engine and to display search results to a user that exceed the confidence probability; and

a display module configured to display the expert search results to the user.

18. The system as in claim 17, further comprising a topic relevance module configured to predict a relevance of a topic in the expert knowledge engine once the search query has been determined to be relevant to a subject of the expert knowledge engine.

19. A system as in claim 17, wherein the predictive model module is constructed using the case library of tagged data and the predictive model is used to determine a probability that a search query is relevant to an expert knowledge engine.

20. A system as in claim 17, wherein the query reformulation module reformulates a query based on expert rules for the expert knowledge engine or topics.