INFORMATION RETRIEVAL USING SUBJECT-AWARE DOCUMENT RANKER

• US 20120197905A1
• Filed: 02/02/2011
• Published: 08/02/2012
• Est. Priority Date: 02/02/2011
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: KUMAR, GIRISH, TAN, ALFIAN, CRASWELL, NICHOLAS ERIC
• Granted Patent: US 8,868,567 B2
• Time in Patent Office: Days
• US Class Current: 707/748
• CPC Class Codes: G06F 16/24578 using ranking
• CPC Class Codes: G06F 16/3344 using natural language analysis
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G06F 40/35 Discourse or dialogue representation

Abstract

Subject matter described herein is related to determining a document score, which suggests a relevance of a document (e.g., webpage) to a search query. For example, a search query is received that is comprised of one or more terms, which represent a subject. An equivalent subject is identified that is semantically similar to the subject. The document score is determined by accounting for both a subject frequency and an equivalent-subject frequency.

Claims

1. Computer-storage media having computer-executable instructions embodied thereon that, when executed, perform a method of determining a document score, which suggests a relevance of a document to a search query, the method comprising:

receiving the search query comprised of one or more terms that represent a subject;

identifying an equivalent subject that is semantically similar to the subject, wherein the subject and the equivalent subject comprise a subject group; and

determining the document score of the document,

wherein the document score is comprised of a subject-group score, and

wherein the subject-group score is calculated using both a subject frequency, which includes a number of times the subject is found in the document, and an equivalent-subject frequency, which includes a number of times the equivalent-subject is found in the document.

2. The computer-storage media of claim 1 further comprising,

parsing the search query into n-gram combinations, wherein the n-gram combinations comprise a plurality of subjects, and

weighting the n-gram combinations to determine a relative importance of each subject among the plurality of subjects.

3. The computer-storage media of claim 2, wherein a weight of an n-gram having more than one search-query term is a function of both a frequency of each term of the n-gram in a corpus and a frequency of the n-gram in the corpus.

4. The computer-storage media of claim 2, wherein a weight of an n-gram is reduced when the n-gram is deemed part of a larger n-gram.

5. The computer-storage media of claim 1, wherein identifying the equivalent subject includes referencing an equivalent-subject datastore, which maintains a listing of equivalent subjects that have been identified as semantically similar to the subject.

6. The computer-storage media of claim 1, wherein, when calculating the subject-group score, the equivalent-subject frequency is weighted based on a confidence that the equivalent subject is semantically similar to the subject.

7. The computer-storage media of claim 1,

wherein the subject-group score is calculated by applying a saturation function to a subject-group frequency, and

wherein the subject-group frequency includes both the subject frequency and the equivalent-subject frequency.

8. The computer-storage media of claim 7, wherein the subject frequency and the equivalent subject frequency are weighted.

9. The computer-storage media of claim 7, wherein the saturation function includes a sum of a customizable parameter and the subject-group frequency.

10. The computer-storage media of claim 1, wherein the document score is calculated using a plurality of subject-group scores that correspond to a plurality of subjects included in the search query.

11. The computer-storage media of claim 11, wherein the document score of the document is comprised of a sum of the plurality of subject-group scores.

12. Computer-storage media having computer-executable instructions embodied thereon that, when executed, perform a method of determining a document score, which suggests a relevance of a document to a search query, the method comprising:

receiving the search query comprised of one or more terms that represent a first subject and a second subject;

identifying a first equivalent subject that is semantically similar to the first subject and a second equivalent subject that is semantically similar to the second subject, wherein each pair of a subject and an equivalent subject comprises a respective subject group;

determining a first-subject-group frequency comprised of both a first-subject frequency, which includes a number of times the first subject is found in the document, and a first-equivalent-subject frequency, which includes a number of times the first equivalent subject is found in the document;

determining a second-subject-group frequency comprised of both a second-subject frequency, which includes a number of times the second subject is found in the document, and a second-equivalent-subject frequency, which includes a number of times the second equivalent subject is found in the document; and

calculating the document score of the document,

wherein the document score is comprised of a first-subject-group score and a second-subject-group score, and

wherein each subject-group score is calculated by applying a saturation function using a respective subject-group frequency.

13. The computer-storage media of claim 12, wherein, when determining the first-subject-group frequency, the first-equivalent-subject frequency is not weighted.

14. The computer-storage media of claim 12, wherein, when determining the first-subject-group frequency, the first-equivalent-subject frequency is weighted based on a confidence that first equivalent subject is semantically similar to the first subject.

15. The computer-storage media of claim 12, wherein the first-subject frequency and the first-equivalent-subject frequency are weighted based on respective locations within the document at which the first subject and first-equivalent subject are found.

16. The computer-storage media of claim 12, wherein identifying an equivalent subject includes referencing an equivalent-subject datastore, which maintains a listing of equivalent subjects that have been identified as semantically similar to the first subject and the second subject.

17. The computer-storage media of claim 12, wherein the saturation function includes dividing the subject-group frequency by a sum of a customizable parameter and the subject-group frequency.

18. The computer-storage media of claim 12, wherein the document score is calculated by adding the first-subject-grouping score and the second-subject-grouping score.

19. A computer system that determines a document score, which suggests a relevance of a document to a search query, the computer system including a processor coupled to computer-storage media, which includes computer software executable by the processor, the computer software comprising:

a search-query receiver that receives the search query, which is comprised of one or more terms that represent a subject;

an equivalent-subject identifier that references an equivalent-subject datastore to identify an equivalent subject, which is semantically similar to the subject, wherein the subject and the equivalent subject comprise a subject group; and

a document ranker that calculates the document score of the document, wherein the document ranker;

(A) determines a subject-group frequency including both a subject frequency, which includes a number of times the subject is found in the document, and an equivalent-subject frequency, which includes a number of times the equivalent subject is found in the document, and

(B) calculates the document score by applying a saturation function using the subject-group frequency.

20. The computer system of claim 19,

wherein, when the saturation function is applied, the subject-group frequency is divided by a sum of a customizable parameter and the subject-group frequency;

wherein applying the saturation function generates a subject-group score; and

wherein the document ranker calculates the document score by combining the subject-group score with another subject-group score of another subject included in the search query.