Retrieving Text from a Corpus of Documents in an Information Handling System

• US 20160055234A1
• Filed: 08/19/2014
• Published: 02/25/2016
• Est. Priority Date: 08/19/2014
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Visotski, William G., Wilson, David E.
• Granted Patent: US 9,727,637 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/3329 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 16/345 Summarisation for human users
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 5/04 Inference or reasoning models
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks

Abstract

A mechanism is provided for retrieving candidate answers from a corpus of documents. The mechanism receives an input question for which an answer is sought. The mechanism extracts features of the input question based on a natural language processing. The mechanism executes a first search of the corpus of documents based on a first subset of the extracted features of the input question and an initial evaluation of a utility of the first subset of extracted features to generate a subset of documents. The mechanism executes a second search of a set of passages extracted from the subset of documents based on a second subset of the extracted features of the input question and a reevaluation of the utility of the second subset of extracted features thereby forming a subset of passages. The mechanism generates query results from the subset of passages matching from which candidate answers are identified.

Claims

1. A method, in a question and answer (QA) system comprising a processor and a memory, for retrieving candidate answers from a corpus of documents, the method comprising:

receiving, by the QA system, an input question for which an answer is sought;

extracting, by the QA system, features of the input question based on a natural language processing of the input question;

executing, by the QA system, a first search of the corpus of documents based on a first subset of the extracted features of the input question and an initial evaluation of a utility of the first subset of extracted features to generate a subset of documents matching the first subset of extracted features;

executing, by the QA system, a second search of a set of passages extracted from the subset of documents based on a second subset of the extracted features of the input question and a reevaluation of the utility of the second subset of extracted features thereby forming a subset of passages; and

generating, by the QA system, query results from the subset of passages from which a set of candidate answers for the input question are identified.

2. The method of claim 1, wherein the set of passages extracted from the subset of documents is less than all of the passages included in the subset of documents.

3. The method of claim 1, wherein executing the first search of the corpus of documents based on the first subset of the extracted features of the input question and the initial evaluation of the utility of the first subset of extracted features to generate the subset of documents matching the first subset of extracted features comprises:

generating, by the QA system, a first statistical data structure for the corpus of documents; and

identifying, by the QA system, the subset of documents from the corpus of documents comprised within the first statistical data structure relevant to the first subset of the extracted features utilizing the initial evaluation of the utility of the first subset of extracted features.

4. The method of claim 1, wherein executing the second search of the set of passages extracted from the subset of documents based on the second subset of the extracted features of the input question and the reevaluation of the utility of the second subset of extracted features comprises:

generating, by the QA system, a second statistical data structure for the set of passages; and

identifying, by the QA system, the query results from the subset of passages comprised within the second statistical data structure relevant to the second subset of the extracted features utilizing the reevaluation of the utility of the second subset of extracted features.

5. The method of claim 1, wherein the extracted features of the input question are identified by:

identifying, by the QA system, a utility of each term in the input question;

eliminating, by the QA system, zero or more terms within the input question that comprise a utility less than a predetermined value; and

adding, by the QA system, the remaining terms in the input question to the extracted features.

6. The method of claim 5, wherein the extracted features of the input question are further identified by:

identifying, by the QA system, one or more synonyms associated with the terms added to the extracted features; and

adding, by the QA system, the one or more synonyms associated with the terms to the extracted features.

7. The method of claim 5, wherein the extracted features of the input question are further identified by:

identifying, by the QA system, one or more tenses associated with the terms added to the extracted features; and

adding, by the QA system, the one or more tenses associated with the terms to the extracted features.

8. A computer program product comprising a computer readable storage medium having a computer readable program stored therein, wherein the computer readable program, when executed on a computing device, causes the computing device to:

receive an input question for which an answer is sought;

extract features of the input question based on a natural language processing of the input question;

execute a first search of a corpus of documents based on a first subset of the extracted features of the input question and an initial evaluation of a utility of the first subset of extracted features to generate a subset of documents matching the first subset of extracted features;

execute a second search of a set of passages extracted from the subset of documents based on a second subset of the extracted features of the input question and a reevaluation of the utility of the second subset of extracted features thereby forming a subset of passages; and

generate query results from the subset of passages from which a set of candidate answers for the input question are identified.

9. The computer program product of claim 8, wherein the set of passages extracted from the subset of documents is less than all of the passages included in the subset of documents.

10. The computer program product of claim 8, wherein the computer readable program to execute the first search of the corpus of documents based on the first subset of the extracted features of the input question and the initial evaluation of the utility of the first subset of extracted features to generate the subset of documents matching the first subset of extracted features further causes the computing device to:

generate a first statistical data structure for the corpus of documents; and

identify the subset of documents from the corpus of documents comprised within the first statistical data structure relevant to the first subset of the extracted features utilizing the initial evaluation of the utility of the first subset of extracted features.

11. The computer program product of claim 8, wherein the computer readable program to execute the second search of the set of passages extracted from the subset of documents based on the second subset of the extracted features of the input question and the reevaluation of the utility of the second subset of extracted features further causes the computing device to:

generate a second statistical data structure for the set of passages; and

identify the query results from the subset of passages comprised within the second statistical data structure relevant to the second subset of the extracted features utilizing the reevaluation of the utility of the second subset of extracted features.

12. The computer program product of claim 8, wherein the extracted features of the input question are identified by the computer readable program causing the computing device to:

identify a utility of each term in the input question;

eliminate zero or more terms within the input question that comprise a utility less than a predetermined value; and

add the remaining terms in the input question to the extracted features.

13. The computer program product of claim 12, wherein the extracted features of the input question are further identified by the computer readable program causing the computing device to:

identify one or more synonyms associated with the terms added to the extracted features; and

add the one or more synonyms associated with the terms to the extracted features.

14. The computer program product of claim 12, wherein the extracted features of the input question are further identified by the computer readable program causing the computing device to:

identify one or more tenses associated with the terms added to the extracted features; and

add the one or more tenses associated with the terms to the extracted features.

15. An apparatus comprising:

a processor; and

a memory coupled to the processor, wherein the memory comprises instructions which, when executed by the processor, cause the processor to;

receive an input question for which an answer is sought;

extract features of the input question based on a natural language processing of the input question;

execute a first search of a corpus of documents based on a first subset of the extracted features of the input question and an initial evaluation of a utility of the first subset of extracted features to generate a subset of documents matching the first subset of extracted features;

execute a second search of a set of passages extracted from the subset of documents based on a second subset of the extracted features of the input question and a reevaluation of the utility of the second subset of extracted features forming a subset of passages; and

generate query results from the subset of passages from which a set of candidate answers for the input question are identified.

16. The apparatus of claim 15, wherein the set of passages extracted from the subset of documents is less than all of the passages included in the subset of documents.

17. The apparatus of claim 15, wherein the instructions to execute the first search of the corpus of documents based on the first subset of the extracted features of the input question and the initial evaluation of the utility of the first subset of extracted features to generate the subset of documents matching the first subset of extracted features further cause the processor:

generate a first statistical data structure for the corpus of documents; and

identify the subset of documents from the corpus of documents comprised within the first statistical data structure relevant to the first subset of the extracted features utilizing the initial evaluation of the utility of the first subset of extracted features.

18. The apparatus of claim 15, wherein the instructions to execute the second search of the set of passages extracted from the subset of documents based on the second subset of the extracted features of the input question and the reevaluation of the utility of the second subset of extracted features further cause the processor to:

generate a second statistical data structure for the set of passages; and

identify the query results from the subset of passages comprised within the second statistical data structure relevant to the second subset of the extracted features utilizing the reevaluation of the utility of the second subset of extracted features.

19. The apparatus of claim 15, wherein the extracted features of the input question are identified by the instructions causing the processor to:

identify a utility of each term in the input question;

eliminate zero or more terms within the input question that comprise a utility less than a predetermined value; and

add the remaining terms in the input question to the extracted features.

20. The apparatus of claim 19, wherein the extracted features of the input question are further identified by the instructions causing the processor to:

identify one or more synonyms associated with the terms added to the extracted features;

add the one or more synonyms associated with the terms to the extracted features;

identify one or more tenses associated with the terms added to the extracted features; and

add the one or more tenses associated with the terms to the extracted features.