Automated opinion prediction based on indirect information

• US 9,672,475 B2
• Filed: 10/14/2015
• Issued: 06/06/2017
• Est. Priority Date: 12/19/2014
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Clark, Adam T., Huebert, Jeffrey K., Payton, Aspen L., Petri, John E.
• Primary Examiner(s): CHEN, ALAN S
• Application Number:
  US14/883,411
• Publication Number:
  US 20160180218A1
• Time in Patent Office: 601 Days
• Field of Search: None
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 5/022 Knowledge engineering; Knowledge acquisition

Abstract

Techniques are disclosed to determine an expected or predicted opinion of a target individual. To do so, a deep question answer system may build a corpus which includes a first collection of documents attributable to a first person and a second collection of documents identified from content in the first collection of documents and evaluate the corpus to build a model representing opinions of the first person relative to topics, concepts, or subjects discussed in the first and second collections of documents. The deep question answer system may also receive a request to predict an opinion of the first person regarding a topic and generate a predicted opinion of the first person regarding the topic from the model.

Claims

1. A computer-implemented method for predicting an opinion, the computer-implemented method comprising:

building a corpus which includes a first collection of documents attributable to a first person and a second collection of documents not attributable to the first person, wherein the second collection of documents is identified from content in the first collection of documents, wherein each of the second collection of documents is available from a respective secondary source;

evaluating the corpus to build a model representing opinions of the first person relative to topics, concepts, or subjects discussed in the first and second collections of documents, including determining, for each secondary source, a source weight factor characterizing a presumed opinion of the secondary source held by the first person;

receiving a request to predict an opinion of the first person regarding a topic specified in the request;

generating a predicted opinion of the first person regarding the topic from the model, by operation of one or more computer processors and based on;

(i) the opinions of the first person relative to topics, concepts, or subjects discussed in the first and second collections of documents and (ii) each source weight factor characterizing the presumed opinion of the respective secondary source held by the first person; and

returning the predicted opinion responsive to the request.

2. The computer-implemented method of claim 1, wherein building the corpus comprises:

parsing the first collection of documents to identify secondary sources, wherein each document in the second collection of documents is available from a respective one of the secondary sources; and

accessing each secondary source to identify documents to add to the second collection of documents.

3. The computer-implemented method of claim 2, wherein evaluating the corpus comprises:

identifying topics, concepts, or subjects referenced by documents in the first collection of documents; and

determining indications of opinions expressed about the topics, concepts, or subjects referenced by documents in the first collection.

4. The computer-implemented method of claim 3, wherein determining the source weight factor comprises correlating the topics, concepts, or subjects identified in one of the secondary sources with the associated topics, concepts, or subjects identified in the first collection of documents.

5. The computed-implemented method of claim 4, wherein the predicted opinion of the first person regarding the topic from the model is generated based on:

(i) indications of opinions expressed about the topics, concepts, or subjects referenced by documents in the first collection;

(ii) indications of opinions on one or more of the topics, concepts, or subjects referenced by documents in the second collection; and

(iii) each source weight factor characterizing the presumed opinion of the respective secondary source held by the first person.

6. The computer-implemented method of claim 1, wherein the first person represents a group of individuals.

7. The computer-implemented method of claim 1, further comprising returning the documents which reference the topic.

8. The computer-implemented method of claim 1, wherein the predicted opinion is generated by an application comprising a plurality of components including a question classifier component, a partial and parallel pipeline execution component, and an evidence parsing component.

9. The computer-implemented method of claim 8, wherein the question classifier component is configured to parse a question in order to determine that the question seeks to predict the opinion and further configured to identify the topic on which the opinion is sought, wherein the topic is stored in a feature store.

10. The computer-implemented method of claim 9, wherein the partial and parallel pipeline execution component is configured to access the identified topic in order to reduce processing of sources that do not discuss the identified topic.

11. The computer-implemented method of claim 10, wherein the evidence parsing component is configured to identify relevant pieces of information from both direct and indirect information, parse the identified information, store the identified information in the corpus, and reprocess sources and topics in the corpus as new sources and topics are identified.

12. The computer-implemented method of claim 11, wherein building the corpus comprises:

parsing the first collection of documents to identify secondary sources, wherein each document in the second collection of documents is available from a respective one of the secondary sources; and

accessing each secondary source to identify documents to add to the second collection of documents.

13. The computer-implemented method of claim 12, wherein evaluating the corpus comprises:

identifying topics, concepts, or subjects referenced by documents in the first collection of documents; and

determining indications of opinions expressed about the topics, concepts, or subjects referenced by documents in the first collection.

14. The computer-implemented method of claim 13, wherein the first person represents a group of individuals, wherein the predicted opinion is returned along with the documents which reference the topic, wherein the predicted opinion is output, wherein evaluating the corpus further includes:

parsing each document accessed from each secondary source to identify topics, concepts, or subjects referenced by the respective document from the respective secondary source; and

determining indications of opinions on the topics, concepts, or subjects referenced by documents in the second collection.

15. The computer-implemented method of claim 1, wherein building the corpus comprises:

parsing the first collection of documents to identify secondary sources.

16. The computer-implemented method of claim 1, wherein evaluating the corpus comprises:

determining indications of opinions expressed about the topics, concepts, or subjects referenced by documents in the first collection.

17. The computer-implemented method of claim 1, wherein the predicted opinion is generated by an application comprising at least one of a question classifier component, a partial and parallel pipeline execution component, and an evidence parsing component.

18. The computer-implemented method of claim 17, wherein the question classifier component is configured to identify the topic on which the opinion is sought.

19. The computer-implemented method of claim 17, wherein the partial and parallel pipeline execution component is configured to reduce processing of sources that do not discuss the identified topic.

20. The computer-implemented method of claim 17, wherein the evidence parsing component is configured to identify relevant pieces of information from at least one of direct information and indirect information.