METHODS AND APPARATUS FOR EVALUATING SEMANTIC PROXIMITY

• US 20100191521A1
• Filed: 04/01/2010
• Published: 07/29/2010
• Est. Priority Date: 06/14/2005
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Colloquis, Inc. (Microsoft Corporation)
• Inventors: Norton, Gray Salmon, Huet, Francois
• Granted Patent: US 7,877,349 B2
• Time in Patent Office: Days
• US Class Current: 704/9
• CPC Class Codes: G06F 16/3344 using natural language analysis
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

Methods and apparatus to evaluate the semantic proximity between reference free-form text entry and a candidate free-form text request.

Claims

1. A system for evaluating a semantic proximity between a question consisting of a string of natural language text presented in a candidate free-form text request received from a user, and a reference free-form text entry consisting of a string of natural language text, comprising:

a data store;

a plurality of reference free-form text entries residing in the data store, each reference free-form text entry in the form of a question expressed as a string of natural language text associated with an answer to the question;

a compiler module, wherein one or more of the reference free-form text entries are received by the compiler module and compiled to form one or more reference patterns corresponding to the reference free-form text entries;

a user interface, wherein a candidate free-form text entry is received in the user interface;

a transformer module, wherein candidate free-form text entries are transformed into candidate rephrased objects; and

a matcher module, wherein the candidate rephrased objects are compared to the output reference patterns and a semantic proximity between the candidate rephrased objects and the reference patterns is calculated and output.

2. A system as in claim 1, wherein the transformer module also transforms reference free-form text entries into reference rephrased objects, and the compiler module further comprises a patternization engine having patternization rules for converting each reference rephrased object into a reference pattern.

3. A system as in claim 2, wherein the transformer module comprises:

a rephrased object engine, wherein a free-form text entry is transformed into a master rephrased object; and

a rephrasing engine, wherein rephrasing rules are applied to the master rephrased object to generate rephrased objects.

4. A system as in claim 3, wherein each rephrasing rule includes a factor expressing how much information to remove from the master rephrased object.

5. A system as in claim 3, wherein each rephrasing rule is coded using a scripting language.

6. A system as in claim 2, wherein the candidate rephrased objects and the reference rephrased objects are data structures comprising:

a representation of information derived from an original free-form text entry; and

a representation of the difference between the information contained in the rephrased object and the original free-form text entry.

7. A system as in claim 6, wherein the data structures further comprise:

information regarding the processing that generated the rephrased object; and

information regarding the context of the original free-form text entry.

8. A system for evaluating a semantic proximity between a question consisting of a string of natural language text presented in a candidate free-form text request received from a user, and a reference free-form text entry consisting of a string of natural language text, comprising:

means for providing a plurality of reference free-form text entries in a data store, each entry in the form of a question expressed as a string of natural language text associated with an answer to the question;

means for receiving a candidate free-form text request from a user work station in the form of a query expressed as a string of natural language text;

means for processing the string of natural language text expressed in each reference free-form text entry into a plurality of reference rephrased objects, each reference rephrased object containing a variation of the string of natural language text expressed in the reference free-form text entry and an expression of the semantic proximity between the variation of the string of natural language text contained in the reference rephrased object and the string of natural language text expressed in the reference free-form text entry;

means for processing the string of natural language text expressed in the candidate free-form text request into a plurality of user rephrased objects, each user rephrased object containing a variation of the string of natural language text expressed in the candidate free-form text request and an expression of the semantic proximity between the variation of the string of natural language text contained in the user rephrased object and the string of natural language text expressed in the candidate free-form text request;

means for matching at least one of the reference rephrased objects derived from the processing of the reference free-form text entry with at least one user rephrased objects derived from the processing of the candidate free-form text request; and

means for calculating a combined semantic proximity for each matching reference rephrased object and user rephrased object.

9. A system as in claim 8, wherein the reference rephrased objects and the user rephrased objects have the same data structure.

10. A system as in claim 8, further comprising means for providing a reference answer to the user based on calculating the combined semantic proximity associated with each matching reference rephrased object and user rephrased object.

11. A system as in claim 8, wherein the expression of the semantic proximity between the variation of the string of natural language text contained in the reference rephrased object and the string of natural language text expressed in the reference free-form text entry is expressed as a percentage.

12. A system as in claim 11, wherein calculating the combined semantic proximity comprises multiplying the percentages of the matched reference rephrased object and user rephrased object.

13. A computer readable storage medium having computer-executable instructions for evaluating a semantic proximity between a question consisting of a string of natural language text presented in a candidate free-form text request received from a user, and a reference free-form text entry consisting of a string of natural language text, comprising:

providing a plurality of reference free-form text entries in a data store, each entry in the form of a question expressed as a string of natural language text associated with an answer to the question;

receiving a candidate free-form text request from a user work station in the form of a query expressed as a string of natural language text;

processing the string of natural language text expressed in each reference free-form text entry into a plurality of reference rephrased objects, each reference rephrased object containing a variation of the string of natural language text expressed in the reference free-form text entry and an expression of the semantic proximity between the variation of the string of natural language text contained in the reference rephrased object and the string of natural language text expressed in the reference free-form text entry;

processing the string of natural language text expressed in the candidate free-form text request into a plurality of user rephrased objects, each user rephrased object containing a variation of the string of natural language text expressed in the candidate free-form text request and an expression of the semantic proximity between the variation of the string of natural language text contained in the user rephrased object and the string of natural language text expressed in the candidate free-form text request;

matching at least one of the reference rephrased objects derived from the processing of the reference free-form text entry with at least one user rephrased objects derived from the processing of the candidate free-form text request; and

calculating a combined semantic proximity for each matching reference rephrased object and user rephrased object.

14. The computer readable storage medium of claim 13, further comprising:

providing a reference answer to the user based on calculating the combined semantic proximity associated with each matching reference rephrased object and user rephrased object.

15. The computer readable storage medium of claim 13, wherein the expression of the semantic proximity between the variation of the string of natural language text contained in the reference rephrased object and the string of natural language text expressed in the reference free-form text entry is expressed as a percentage.

16. The computer readable storage medium of claim 15, wherein the step of calculating the combined semantic proximity comprises multiplying the percentages of the matched reference rephrased object and user rephrased object.

17. The computer readable storage medium of claim 13, wherein the reference rephrased objects and the user rephrased objects use the same data structure.

18. The computer readable storage medium of claim 13, wherein the processing applied to convert a reference free-form text entry into the list of reference rephrased objects is the same as the processing applied to convert a candidate free-form text request into a list of user rephrased objects.