Using cohorts to infer attributes for an input case in a question answering system

• US 9,818,062 B2
• Filed: 02/19/2015
• Issued: 11/14/2017
• Est. Priority Date: 02/19/2015
• 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., Stevens, Richard J., Whaley, Kathryn L.
• Primary Examiner(s): Wong, Lut
• Application Number:
  US14/626,052
• Publication Number:
  US 20160247071A1
• Time in Patent Office: 999 Days
• Field of Search: 706 11
• CPC Class Codes: G06F 16/24522 Translation of natural language queries to structured queries
• CPC Class Codes: G06F 16/90324 using system suggestions
• CPC Class Codes: G06F 16/90332 Natural language query formulation or dialogue systems
• CPC Class Codes: G06N 5/04 Inference or reasoning models

Abstract

A cohort analysis mechanism analyzes cohorts to infer one or more additional attributes for an input case to provide a refined input case to answer a question in a question answering system. The refined input case is then used to answer a question in the question answering system. The refined input case can be used in a traditional question answering flow or in a flow that again uses cohort analysis to extract relevant data to answer the question. The cohort analysis mechanism analyzes cohorts to find common attributes in the cohorts and then determines whether to infer the common attributes into the refined input case. The cohort analysis mechanism may determine to dialog with a user to confirm an inferred attribute.

Claims

1. A method for answering a question in a question answering system, the method comprising the steps of:

analyzing an input case representing the question to determine whether cohorts can be used with the input case, wherein the cohorts are similar entities with similar characteristics to an entity of the input case and available in a corpus of data;

using cohort analysis to infer an attribute for the input case comprising;

determining if the input case has an entity that has available cohorts;

identifying cohorts in the corpus of data using fuzzy matching with a threshold percentage and where the cohorts are similar to the entity in the input case;

analyzing the cohorts to find common attributes in the cohorts;

finding missing attributes for the entity of the input case from the common attributes in the cohorts to infer a missing attribute for the input case;

scoring the missing attributes according to consistency and sample size;

applying confidence thresholds to the missing attributes to determine whether to infer the attribute for the refined input case;

determining whether to dialog with a user to confirm the inferred attribute using a lower threshold compared to inferring the attribute without dialoging with the user;

generating a refined input case that includes the inferred attribute applied to the entity of the input case; and

using the refined input case with the inferred attribute to answer the question.

2. The method of claim 1 wherein the cohorts are selected from patients, people, animals and computer components.

3. The method of claim 1 wherein the input case is a medical question for a patient and the cohort analysis mechanism processes the medical history of the patient and finds cohorts, wherein the cohorts comprise a set of other patients with similar attributes, by processing medical histories of the other patients to infer the attribute for the refined input case from the cohorts.

4. The method of claim 1 wherein the cohort analysis mechanism is part of a question answering application that answers a natural language question.

5. A method for utilizing cohorts in a question answering system, the method comprising:

analyzing an input case in a question answering application that answers a natural language question to determine whether cohorts can be used with the input case, wherein the cohorts are similar entities with similar characteristics to an entity of the input case and available in a corpus of data;

using cohort analysis to infer an attribute for the input case comprising;

determining if the input case has an entity that has available cohorts;

identifying cohorts in the corpus of data using fuzzy matching with a threshold percentage and where the cohorts are similar to the entity in the input case;

analyzing the cohorts to find common attributes in the cohorts and finding missing attributes for the entity of the input case for the common attributes in the cohorts to infer a missing attribute for the input case;

scoring the missing attributes according to consistency and sample size;

applying confidence thresholds to the missing attributes to determine whether to infer the attribute for the refined input case; and

determining whether to dialog with a user to confirm the inferred attribute using a lower threshold compared to inferring the attribute without dialoging with the user;

providing a refined input case that includes the inferred attribute; and

answering a question with the refined input case.

6. The method of claim 5 wherein the input case is a medical question for a patient and the cohort analysis mechanism processes the medical history of the patient and finds cohorts, wherein the cohorts comprise a set of other patients with similar attributes, by processing medical histories of the other patients to infer the attribute for the refined input case from the cohorts.