Web search ranking

Patent metadata

Field	Value
Publication number	US-9104733-B2
Application number	US-201213688209-A
Country	US
Kind code	B2
Filing date	Nov 29, 2012
Priority date	Nov 29, 2012
Publication date	Aug 11, 2015
Grant date	Aug 11, 2015

Abstract

A computer-implemented method and system for Web search ranking are provided herein. The method includes generating a number of training samples from clickthrough data, wherein the training samples include positive query-document pairs and negative query-document pairs. The method also includes discriminatively training a translation model based on the training samples and ranking a number of documents for a Web search based on the translation model.

First claim

What is claimed is:

1. A computer-implemented method for Web search ranking, comprising:
generating a plurality of training samples from clickthrough data, wherein the plurality of training samples comprises positive query-document pairs and negative query-document pairs, and wherein position bias from the clickthrough data is removed during generation of the plurality of training samples;
discriminatively training a translation model based on the plurality of training samples; and
ranking a plurality of documents for a Web search based on the translation model.

2. A system for Web search ranking, comprising:
a processor that is adapted to execute stored instructions; and
a system memory, wherein the system memory comprises code configured to:
generate a plurality of training samples from clickthrough data, wherein position bias within the clickthrough data is removed during generation of the plurality of training samples;
discriminatively train a sparse log-linear translation model based on the plurality of training samples; and
rank a plurality of documents for a Web search based on the sparse log-linear translation model.

3. The system of claim 2 , wherein ranking the plurality of documents for the Web search comprises assigning a relevance score to each of the plurality of documents based on a relevance of the document to a query.

4. The system of claim 2 , wherein the plurality of training samples comprises positive query-document pairs and negative query-document pairs.

5. The system of claim 4 , wherein each of the positive query-document pairs and each of the negative query-document pairs is generated based on a similarity between terms in a query and terms in a specified field of a document corresponding to the query.

6. The system of claim 5 , wherein the specified field comprises a title field of the document.

7. The system of claim 2 , wherein the clickthrough data comprise user queries, documents corresponding to the user queries, and user click information corresponding to the user queries.

8. The system of claim 2 , wherein the sparse log-linear translation model is discriminatively trained using an L1-norm regularization technique.

9. The system of claim 8 , wherein the L1-norm regularization technique comprises an Orthant-Wise Limited-memory Quasi-Newton (OWL-QN) technique.

10. The system of claim 2 , wherein removing the position bias comprises assigning a weight to each query-document pair based on a location of a corresponding document within a search result list of a corresponding query, and wherein documents that are higher in the search result list are assigned a lower weight.

11. One or more computer-readable storage media for storing computer-readable instructions, the computer-readable instructions providing a Web search ranking system when executed by one or more processing devices, the computer-readable instructions comprising code configured to:
generate a plurality of training samples from clickthrough data, wherein the plurality of training samples comprises positive query-document pairs and negative query-document pairs, and wherein position bias within the clickthrough data is removed during generation of the plurality of training samples;
discriminatively train a translation model based on the plurality of training samples; and
rank a plurality of documents for a Web search based on the translation model.

12. The one or more computer-readable storage media of claim 11 , wherein ranking the plurality of documents for the Web search comprises assigning a relevance score to each of the plurality of documents based on a relevance of the document to a query.

13. The one or more computer-readable storage media of claim 11 , wherein the translation model is discriminatively trained using an L1-norm regularization technique.

14. The one or more computer-readable storage media of claim 13 , wherein the L1-norm regularization technique comprises an Orthant-Wise Limited-memory Quasi-Newton (OWL-QN) technique.

Assignees

• Microsoft Corp
• Microsoft Technology Licensing Llc

Inventors

• Gao Jianfeng
• Qu Zhonghua
• Xu Gu

Classifications

• G06F17/30702

  Physics · mapped topic

• G06F17/3053 Primary

  Physics · mapped topic

• G06F17/30864

  Physics · mapped topic

• G06F16/337 Primary

  Profile generation, learning or modification · CPC title

• G06F16/951

  Indexing; Web crawling techniques · CPC title