Providing answers to questions having both rankable and probabilistic components

• US 9,940,355 B2
• Filed: 09/23/2015
• Issued: 04/10/2018
• Est. Priority Date: 03/09/2015
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Prager, John Martin
• Primary Examiner(s): CHEN, ALAN S
• Application Number:
  US14/862,543
• Publication Number:
  US 20160267392A1
• Time in Patent Office: 930 Days
• Field of Search: None
• CPC Class Codes: G06F 16/24522 Translation of natural language queries to structured queries
• CPC Class Codes: G06F 16/3329 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 16/3346 using probabilistic model
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G06F 40/40 Processing or translation of natural language natural language analysis G06F40/20; semantic analysis G06F40/30
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks

Abstract

Providing answers to questions. A LAT, a rankable criterion, and a probabilistic criterion are identified in a received query. A set of candidate answers to the question that are instances of the LAT identified in a first data source is generated. Based on the rankable criterion, the candidate answers are ranked. Each candidate answer is assigned a likelihood that the candidate answer satisfies the probabilistic criterion, based on a statistic of occurrences of terms related to the candidate answer and the probabilistic criterion in text passages from a second data source. One or more candidate answers are selected based on the rank and the likelihood of the candidate answers. In another aspect of the invention, a Boolean criterion is also identified in the query and the set of candidate answers is reduced by applying the Boolean criterion.

Claims

1. A computer-implemented method for providing answers to questions, the method comprising:

receiving, by a computer, a query;

identifying, by the computer, a lexical answer type (LAT), a rankable criterion, and a probabilistic criterion in the query, wherein a rankable criterion refers to components of the query that contain ordinals and/or superlatives, and a probabilistic criterion refers to components of the query that contain phrases that modify the LAT;

generating, by the computer, a set of candidate answers to the query that are instances of the LAT identified in a structured data source;

assigning, by the computer, a rank to each candidate answer, based on the rankable criterion;

assigning, by the computer, a likelihood to each candidate answer, that the candidate answer satisfies the probabilistic criterion, based on a statistic of occurrences of terms related to the candidate answer and to the probabilistic criterion in text passages from an unstructured data source;

selecting, by the computer, one or more candidate answers, based on the product of the likelihood that the candidate answer satisfies the probabilistic criterion and the likelihoods that each candidate answer of lower rank does not satisfy the probabilistic criterion; and

transmitting, by the computer, the selected candidate answers.

2. A method in accordance to claim 1, wherein

identifying, by the computer, further comprises identifying a Boolean criterion in the query, wherein a Boolean criterion refers to a component of the query that modifies the LAT and is either entirely true or entirely false for the LAT, and the probabilistic criterion refers to components of the query that contain phrases that modify the LAT and are not Boolean criterion;

and

wherein the method further comprises;

reducing, by the computer, the set of candidate answers by applying the Boolean criterion.

3. A method in accordance to claim 1, wherein identifying, by the computer, further comprises using a natural language parser or a named entity recognition program.

4. A method in accordance to claim 1, further comprising:

identifying a subset of the ranked candidate answers based on a threshold value or filter; and

wherein assigning a likelihood further comprises;

assigning, by the computer, to each ranked candidate answer in the identified subset of ranked candidate answers a likelihood that the ranked candidate answer satisfies the probabilistic criterion, based on a statistic of occurrences of terms related to the ranked candidate answer and to the probabilistic criterion in text passages from an unstructured data source;

and wherein selecting further comprises;

selecting, by the computer, one or more candidate answers in the identified subset of ranked candidate answers, based on the rank and the likelihood of the candidate answers.