Identifying Query Intent

• US 20150324440A1
• Filed: 05/12/2014
• Published: 11/12/2015
• Est. Priority Date: 05/12/2014
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Inventors: Subramanian, Meyyammai, Hassan, Sayed, Wirth, Anneliese Creighton, Valeriano, Michelle Follosco
• Granted Patent: US 9,934,306 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/24522 Translation of natural language queries to structured queries
• CPC Class Codes: G06F 16/24564 Applying rules; Deductive queries
• CPC Class Codes: G06F 16/285 Clustering or classification
• CPC Class Codes: G06F 16/3338 Query expansion
• CPC Class Codes: G06F 16/951 Indexing; Web crawling techniques
• CPC Class Codes: G06F 9/453 Help systems

Abstract

Technologies are described herein for identifying query intent from a raw query. A method for identifying intent may include repeatedly separating and merging terms of a natural language expression based on a set of rule-based transpositions of natural language terms into one or more defined terms based on predetermined naming conventions for at least one software function. Thereafter, a cluster of previous search terms related to the defined terms may be identified, and the natural language expression may be associated with the identified cluster to create intent-based cluster information.

Claims

1. A computer-implemented method for identifying query intent, the method comprising:

receiving a raw query directed to at least one function provided by a software product available at a computer, the raw query comprising a natural language expression related to the at least one function;

repeatedly separating and merging terms of the natural language expression based on a set of rule-based transpositions of natural language terms into one or more defined terms based on predetermined naming conventions for the at least one function, the one or more defined terms referring to the at least one function;

identifying a cluster of previous search terms related to the one or more defined terms; and

associating the natural language expression to the identified cluster to create a set of intent-based cluster information, wherein the query intent is represented by the one or more defined terms.

2. The computer-implemented method of claim 1, wherein repeatedly separating and merging terms of the natural language expression comprises:

grouping terms of the natural language expression into a set of preliminary groupings of terms based on modifying terms in the natural language expression, the modifying terms referring to a task using the at least one function and the set of rule-based transpositions being applied to the preliminary groupings of terms.

3. The computer-implemented method of claim 2, wherein repeatedly separating and merging terms of the natural language expression further comprises:

repeatedly separating and merging individual groupings of the set of preliminary groupings based on the set of rule-based transpositions of natural language terms into the one or more defined terms.

4. The computer-implemented method of claim 1, wherein the set of rule-based transpositions comprise at least one of:

a rule to transpose one or more terms of the natural language expression into synonyms related to the at least one function;

a rule to transpose one or more terms of the natural language expression into clauses related to the at least one function;

a rule to transpose one or more terms of the natural language expression into root lemmas of the one or more terms; and

a rule to transpose one or more terms of the natural language expression into phrases contained in the cluster of previous search terms.

5. The computer-implemented method of claim 1, wherein the set of rule-based transpositions comprise at least one of:

a rule to group at least two terms of the natural language expression as a token referring to an aspect of the at least one function; and

a rule to separate a phrase of the natural language expression into one or more terms based on identification of more than one verb in the phrase.

6. The computer-implemented method of claim 1, wherein the set of rule-based transpositions comprise at least one of:

a rule to group at least two terms of the natural language expression as a phrase referring to a portion of a description of the at least one function; and

a rule to remove at least one term from the natural language expression to create a phrase more closely related to the at least one function.

7. The computer-implemented method of claim 1, wherein the set of rule-based transpositions comprise at least one of:

a rule to identify parts of speech of the natural language expression; and

a rule to override identified parts of speech to create a phrase more closely related to the at least one function.

8. The computer-implemented method of claim 1, wherein the set of rule-based transpositions comprise a rule to override corrected spellings of terms to create a phrase more closely related to the at least one function.

9. The computer-implemented method of claim 1, wherein identifying the cluster of previous search terms related to the one or more defined terms comprises comparing the one or more defined terms to a plurality of cluster identification information to determine a matching cluster.

10. The computer-implemented method of claim 1, wherein associating the natural language expression to the identified cluster comprises associating at least one phrase of the natural language expression to the identified cluster to create a one-to-one representation for matching of new raw queries containing the at least one phrase to the identified cluster.

11. A computer-readable storage medium having computer executable instructions stored thereon which, when executed by a computer, cause the computer to:

receive a raw query directed to at least one function provided by a software product available at a computer, the raw query comprising a natural language expression related to the at least one function and having an intent to utilize or identify the at least one function;

group terms of the natural language expression into a set of preliminary groupings of terms based on modifying terms in the natural language expression, the modifying terms referring to a task using the at least one function;

separate and merge individual groupings of the set of preliminary groupings based on rule-based transpositions of natural language terms into refined groupings of terms based on predetermined naming conventions for the at least one function, the refined groupings of terms referring to the at least one function;

identify a cluster of previous search terms related to the refined groupings of terms; and

associate the natural language expression to the identified cluster to create a set of intent-based cluster information, wherein the intent is represented by the refined groupings of terms.

12. The computer-readable storage medium of claim 11, wherein the set of rule-based transpositions comprise at least one of:

a rule to transpose one or more terms of the natural language expression into a grouping of synonyms related to the at least one function;

a rule to transpose one or more terms of the natural language expression into a grouping of clauses related to the at least one function;

a rule to transpose one or more terms of a grouping into root lemmas of the one or more terms; and

a rule to transpose one or more terms of the natural language expression into a grouping contained in the cluster of previous search terms.

13. The computer-readable storage medium of claim 11, wherein the set of rule-based transpositions comprise at least one of:

a rule to group at least two terms of the natural language expression as a token grouping referring to an aspect of the at least one function; and

a rule to separate a grouping into one or more smaller groupings based on identification of more than one verb in the grouping.

14. The computer-readable storage medium of claim 11, wherein the set of rule-based transpositions comprise at least one of:

a rule to group at least two terms of the natural language expression as a grouping referring to a portion of a description of the at least one function; and

a rule to remove at least one term from another grouping to create a grouping more closely related to the at least one function.

15. The computer-readable storage medium of claim 11, wherein the set of rule-based transpositions comprise at least one of:

a rule to identify parts of speech of the natural language expression; and

a rule to override identified parts of speech to create a grouping more closely related to the at least one function.

16. The computer-readable storage medium of claim 11, wherein the set of rule-based transpositions comprise a rule to override corrected spellings of terms in the groupings to create a grouping more closely related to the at least one function.

17. The computer-readable storage medium of claim 11, wherein identify the cluster of previous search terms related to the refined groupings of terms comprises comparing the refined groupings of terms to a plurality of cluster identification information to determine a matching cluster.

18. The computer-readable storage medium of claim 11, wherein associate the natural language expression to the identified cluster comprises associating at least one grouping of the refined groupings to the identified cluster to create a one-to-one representation for matching of new raw queries containing terms of the at least one grouping to the identified cluster.

19. The computer-readable storage medium of claim 11, having instructions to further cause the computer to:

receive a new raw query containing at least one phrase;

compare the at least one phrase to the intent-based cluster information to identify an associated cluster of previous search terms; and

returning the identified associated cluster as a result for the new raw query.

20. A clustering system configured to identify intent from raw queries received from a plurality of client computers directed to at least one function provided by a software product, the system comprising:

at least one computer executing a raw query processing service configured to receive a raw query from at least one client computer of the plurality of client computers,

correct spelling of words included in the raw query to create a spell-corrected query,

override corrected spellings of a portion of the words in the spell-corrected query to create an initial grouping related to the at least one function,

group terms of the initial grouping and the spell-corrected query into a set of preliminary groupings of terms based on modifying terms in the natural language expression, the modifying terms referring to a task using the at least one function,

separate and merge individual groupings of the set of preliminary groupings based on rule-based transpositions of natural language terms into refined groupings of terms based on predetermined naming conventions for the at least one function, the refined groupings of terms referring to the at least one function,

identify a cluster of previous search terms related to the refined groupings of terms,

associate the natural language expression to the identified cluster to create a set of intent-based cluster information, wherein the intent is represented by the refined groupings of terms,

receive a new raw query containing at least one phrase, and

compare the at least one phrase to the intent-based cluster information to identify an associated cluster of previous search terms.