GENERATING SYNONYMS BASED ON QUERY LOG DATA

• US 20100082657A1
• Filed: 09/23/2008
• Published: 04/01/2010
• Est. Priority Date: 09/23/2008
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Cheng, Tao, Lauw, Hady W., Paparizos, Stelios
• Granted Patent: US 9,092,517 B2
• Time in Patent Office: Days
• US Class Current: 707/767
• CPC Class Codes: G06F 16/374 Thesaurus

Abstract

An approach is described for generating synonyms to supplement at least one information item, such as, in one case, a set of related items. The approach can involve an expansion phase, a clean-up phase, and a reduction phase. In the expansion phase, the approach identifies, for each related item, a set of initial synonym candidates. In the clean-up phase, the approach removes noise from the set of initial synonym candidates (if such noise exists), to provide a set of filtered synonym candidate items. In the reduction phase, the approach ranks and applies a threshold (or thresholds) to the set of filtered synonym candidate items, to generate, for each information item, a set of selected synonyms. The approach uses query log data as at various points in its operation. The selected synonyms can be used to improve the effectiveness of user searches.

Claims

1. A method for generating synonyms using data processing functionality, comprising:

receiving a set of input information items;

for each information item in the set of input information items;

generating a set of initial synonym candidates based on query log data;

removing noise from the set of initial synonym candidates, if said noise is present and can be identified, an output of said generating and removing comprising a set of filtered synonym candidates; and

reducing the set of filtered synonym candidates to a set of selected synonyms; and

outputting synonym-expanded data that is formed based on said reducing performed with respect to each information item in the set of input information items.

2. The method of claim 1, wherein the set of input information items comprises a list of canonical information items associated with a common theme.

3. The method of claim 1, wherein the set of input information items comprises at least part of a collection of data, the collection of data being at least partially structured.

4. The method of claim 1, wherein the set of input information items comprises a single information item, and wherein no noise is removed from the set of initial synonym candidates, making the set of initial synonym candidates the same as the set of filtered synonym candidates.

5. The method of claim 1, wherein said generating comprises:

identifying, for each information item, a set of k associated page items;

for each page item in the set of k page items, identifying, based on the query log data, a set of associated queries; and

providing, for each information item, the set of initial synonym candidates by combining sets of queries identified for respective page items in the set of k page items.

6. The method of claim 5, wherein said identifying a set of k page items comprises:

submitting each information item to a search module;

receiving, for each information item, a set of associated initial page items; and

selecting, for each information item, the k page items from the set of associated initial page items.

7. The method of claim 1, wherein said removing comprises:

identifying, for each information item, at least one context-sensitive noise component within the set of initial synonym candidates, if said at least one context-sensitive noise component is present and can be identified, said at least one context-sensitive noise component affecting synonym candidates associated with two or more information items in the set of input information items; and

removing said at least one context-sensitive noise component from the set of initial synonym candidates.

8. The method of claim 1, wherein said reducing comprises:

obtaining at least one metric factor associated with the set of filtered synonym candidates for each information item, based on the query log data;

ranking the set of filtered synonym candidates based on said at least one metric factor to provide a set of ranked synonym candidates; and

selecting the set of selected synonyms from the set of ranked synonym candidates based on said at least one metric factor.

9. The method of claim 8, wherein said selecting of the set of selected synonyms comprises applying a threshold to the set of ranked synonym candidates to identify the set of selected synonyms.

10. The method of claim 1, wherein the synonym-expanded data comprises the set of input information items in canonical form, supplemented to include, for each input information item, an associated set of selected synonyms.

11. A synonym-generating module, comprising:

an input module configured to receive a set of input information items;

an expansion module configured to expand each information item in the set of input information items into a set of initial synonym candidates based on query log data, the query log data reflecting associations between prior queries submitted by users and page items accessed by the users in response to the submitted queries;

a clean-up module configured to remove, for each information item, noise from the set of initial synonym candidates, if said noise is present and can be identified, an output of the expansion module and the clean-up module comprising a set of filtered synonym candidates; and

a reduction module configured to select, for each information item, a set of synonyms from the set of filtered synonym candidates based on the query log data, to provide a set of selected synonyms; and

an output module configured to output synonym-expanded data that is formed based on the selecting performed by the reduction module with respect to each information item in the set of input information items.

12. The synonym-generating module of claim 11, wherein the set of input information items comprises a list of canonical information items associated with a common theme.

13. The synonym-generating module of claim 11, wherein the set of input information items comprises a single information item, and wherein no noise is removed from the set of initial synonym candidates by the clean-up module, making the set of initial synonym candidates the same as the set of filtered synonym candidates.

14. The synonym-generating module of claim 11, wherein the expansion module comprises:

logic configured to identify, for each information item, a set of k associated page items;

logic configured to identify, for each page item in the set of k page items, a set of associated queries based on the query log data; and

logic configured to provide, for each information item, the set of initial synonym candidates by combining sets of queries identified for respective page items in the set of k page items.

15. The synonym-generating module of claim 14, wherein said logic configured to identify a set of k page items comprises:

log configured to submit each information item to a search module;

logic configured to receive, for each information item, a set of associated initial page items; and

logic configured to select, for each information item, the k page items from the set of associated initial page items.

16. The synonym-generating module of claim 11, wherein said clean-up module comprises:

logic configured to identify, for each information item, at least one context-sensitive noise component within the set of initial synonym candidates, if said at least one context-sensitive noise component is present and can be identified, said at least one context-sensitive noise component affecting synonym candidates associated with two or more information items in the set of input information items; and

logic configured to remove said at least one context-sensitive noise component from the set of initial synonym candidates.

17. The synonym-generating module of claim 11, wherein said reduction module comprises:

logic configured to obtain at least one metric factor associated with the set of filtered synonym candidates for each information item, based on the query log data;

logic configured to rank the set of filtered synonym candidates based on said at least one metric factor to provide a set of ranked synonym candidates; and

logic configured to select the set of selected synonyms from the set of ranked synonym candidates based on said at least one metric factor.

18. The synonym-generating module of claim 17, wherein said logic configured to select the set of selected synonyms comprises logic configured to apply a threshold to the set of ranked synonym candidates to identify the set of selected synonyms.

19. The synonym-generating module of claim 11, wherein the synonym-expanded data comprises the set of input information items in canonical form, supplemented to include, for each input information item, an associated set of selected synonyms.

20. A computer-readable medium for storing computer-readable instructions, the computer-readable instructions providing a synonym-generating module when executed by one or more processing devices, the computer-readable instructions comprising:

logic configured to receive at least partially structured data, said at least partially structured data including a set of input information items, the set of input information items including one or more information items; and

logic configured to generate, for each information item in the set of input information items, a set of selected synonyms based on query log data, the query log data reflecting associations between prior queries submitted by users and page items accessed by the users in response to the submitted queries.