QUERY CLASSIFICATION

• US 20100293174A1
• Filed: 05/12/2009
• Published: 11/18/2010
• Est. Priority Date: 05/12/2009
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Colllins-Thompson, Kevyn B., Chickering, David Maxwell, Liebling, Daniel J., Bennett, Paul N., Dumais, Susan
• Granted Patent: US 8,719,249 B2
• Time in Patent Office: Days
• US Class Current: 707/759
• CPC Class Codes: G06F 16/353 into predefined classes

Abstract

Techniques and systems are disclosed that provide for constructing a query classification index that can be used to classify a query into relevant categories. Where documents in an index are classified into one or more category predictions for a category hierarchy, classification metadata is generated for categories to which a document in the index has been classified. Further, the classification metadata is associated to the corresponding documents in the index. Additionally, a query of the index can be classified using the metadata associated to the documents in the index, and query results can be provided that are classified by the one or more categories identified by the classification of the query.

Claims

1. A method for providing query results based on categorization of searched content, comprising:

generating classification metadata that identify a category to which a document in an index of documents is classified, where documents in the index have been classified into one or more category predictions for a category hierarchy; and

associating the classification metadata to the document in the index.

2. The method of claim 1, comprising:

classifying a query of the index using the metadata associated to the documents in the index; and

providing query results classified by one or more categories using the classification of the query.

3. The method of claim 1, classifying documents in an index into category predictions for a category hierarchy, comprising one or more of:

using a trained classifier to identify one or more categories for the document, based on a desired threshold; and

determining a retrieval relevance score to identify one or more categories for the document, based on a desired threshold.

4. The method of claim 2, classifying a query of the index using the metadata associated to the documents in the index, comprising:

identifying document results for the query from the index;

identifying one or more categories for respective documents from the results using a distribution of the categories for the results provided by the metadata; and

providing one or more categories for the query based on a distribution of the one or more identified categories for respective documents in the results.

5. The method of claim 1, comprising building a mapping of query terms to a category distribution for the index, comprising:

mapping respective terms in a document in the index to the one or more category predictions for the document;

aggregating the terms mapped for the index to create a distribution of category predictions for the respective terms.

6. The method of claim 5, classifying a query of the index using the metadata associated to the documents in the index, comprising:

mapping respective query terms to the distribution of category predictions for the respective terms in the index;

identifying one or more categories for the query using the query term mapping; and

providing one or more predicted relevant topic models for the query based on the distribution of the identified categories for the query.

7. The method of claim 2, comprising determining ambiguity for the query comprising using an ambiguity function that accounts for a spread of the distribution of the categories associated with the query.

8. The method of claim 7, determining an ambiguity score for the query comprising measuring the spread of the distribution of a desired number of categories associated with the query.

9. The method of claim 2, comprising determining a query similarity between the query and an alternate query comprising comparing the distribution of the categories associated with the query and the distribution of the categories associated with the alternate query.

10. The method of claim 9, determining a query similarity score for the query with respect to the alternate query comprising calculating a similarity function using the distribution of the categories associated with the query and the distribution of the categories associated with the alternate query.

11. The method of claim 2, comprising determining a query quality characteristic, comprising combining the determination of query ambiguity with the determination of query similarity between the query and an alternate query.

12. The method of claim 2, comprising determining a query refinement score that is used to measure whether an alternate query is one or more of:

a more specific refinement of one or more portions of the query;

a more general refinement of one or more portions of the query; and

an orthogonal concept to one or more portions of the query;

13. The method of claim 1, where a query comprises one or more of:

keyword terms from a search-based query;

keywords associated with an Internet-based advertisement;

tags terms associated with a file; and

keywords associated with a caption.

14. The method of claim 1, classifying an index of documents comprising classifying a federation of documents from different source collections.

15. The method of claim 1, comprising one or more of:

generating classification metadata at crawl time of a source of documents for the index; and

associating the classification metadata to the document at crawl time of a source of documents for the index.

16. A system for providing query results based on categorization of searched content, comprising:

computer-based storage configured to store classification metadata associated with documents in an index of documents;

a metadata generator configured to generate classification metadata that identify a category to which a document in the index has been classified, where documents in the index have been classified into one or more category predictions for a category hierarchy; and

a metadata association component, operably coupled to the storage, and configured to associate the classification metadata to the corresponding document in the index.

17. The system of claim 16, comprising:

a query classifier configured to classify a query of the index using the metadata associated to the documents in the index comprising a query distribution component configured to generate a distribution of categories, provided by the metadata, for query results; and

a classified query results provider operably coupled to the query classifier and configured to provide query results classified by one or more categories using the classification of the query from the query classifier.

18. The system of claim 16 comprising a trained classifier configured to identify one or more categories for a document in the index, based on a desired threshold.

19. The system of claim 16, comprising:

a document term mapper configured to;

map respective terms in a document in the index to the one or more category predictions for the document; and

aggregate the terms mapped for the index to create a distribution of category predictions for the respective terms; and

a mapped query classifier configured to;

map respective query terms, for a query to the index, to the distribution of category predictions for the respective terms in the index;

identify one or more categories for the query using the query term mapping; and

provide one or more predicted relevant topic models for the query based on the distribution of the identified categories for the query.

20. A method for providing query results based on categorization of searched content, comprising:

classifying document in an index into topic category prediction for a category hierarchy, comprising using a trained classifier to identify one or more categories for the document, based on a desired threshold;

generating classification metadata, at crawl time of a source of documents for the index, that identify a topic category to which a document has been classified;

associating the classification metadata to the document in the index at crawl time of a source of documents for the index;

classifying a query of the index using the metadata associated to the documents in the index, comprising;

identifying document results for the query from the index; and

identifying one or more categories for respective documents from the results using a distribution of the categories for the results provided by the metadata; and

providing one or more categories for the query based on a distribution of the one or more identified categories for respective documents in the results and

providing query results classified by one or more categories using the classification of the query.