US 10,255,273 B2Granted

Method and system for ranking and summarizing natural language passages

AssigneeMicrosoft Technology Licensing, LLC

InventorsDoran Chakraborty, Manish Malik, Qifa Ke, Jonathan R. Tiao

Life of this patent

Estimating the term…

Loading patent…

Abstract

Examples of the present disclosure describe systems and methods relating to generating a relevance score on a given natural language answer to a natural language query for ranking the answer among other answers for the query, while generating a summary passage and a likely query to the given passage. For instance, multi-layered, recurrent neural networks may be used to encode the query and the passage, along with a multi-layered neural network for information retrieval features, to generate a relevant score for the passage. A multi-layered, recurrent neural network with soft attention and sequence-to-sequence learning task may be used as a decoder to generate a summary passage. A common encoding neural network may be employed to encode the passage for the ranking and the summarizing, in order to present concise and accurate natural language answers to the query.

Claims20

Dependent claims

A system comprising:

at least one processor;
and
a memory encoding computer executable instructions that, when executed by the at least one processor, perform a method for generating a relevance score on a natural language candidate answer to a natural language query for one or more electronic files, the method comprising:
generating a first semantic vector expression, wherein the first semantic vector represents data for semantic features of the natural language query;
generating a second semantic vector expression, wherein the second semantic vector represents data for semantics features of the natural language candidate answer;
generating a combined semantic vector expression based on at least one dimension of each of the first semantic vector expression and the second semantic vector expression using one or more multi-layered recurrent neural networks, wherein the combined semantic vector expression represents a data structure for semantic characteristics of the natural language query and the natural language candidate answer;
generating a textual vector expression based on a pair of the natural language query and the natural language candidate answer using a multi-layered neural network as textual analyzer, wherein the textual vector expression represents textual features of the pair of the natural language query and the natural language candidate answer, wherein the textual features comprise a degree of matching between the pair based on appearance of words;
generating a relevance score based on the combined semantic vector expression and the textual vector expression using a relevance score generator based on a weighted combination of coefficient values of at least one dimension of the combined semantic vector expression and the textual vector expression;
and
providing the relevance score for ranking the natural language candidate answer to the natural language query.

7 more elements

A system comprising:

at least one processor;
and
a memory encoding computer executable instructions that, when executed by the at least one processor, perform a method for generating a relevance score on a natural language candidate answer to a natural language query for one or more electronic files, the method comprising:
receiving a natural language query;
generating a first semantic vector expression based on a natural language query using a first encoder;
generating a second semantic vector expression based on the natural language candidate answer using a second encoder;
generating a textual vector expression based on a pair of the natural language query and the natural language candidate answer using a textual analysis system, wherein the textual vector expression represents textual characteristics of a pair of the natural language query and the natural language candidate answer, wherein the textual characteristics comprise a degree of matching between the pair based on appearance of words;
generating a third semantic vector expression based on the first semantic vector expression and the second semantic vector expression using a comparator;
generating a relevance score based on the third semantic vector expression and the textual vector expression using a relevance score generator;
and
providing the relevance score for ranking the natural language candidate answer to the natural language query.

8 more elements

A computer-implemented method comprising:

generating a combined semantic vector expression based on at least one dimension of each of the first semantic vector and the second semantic vector using one or more multi-layered recurrent neural networks, wherein the combined semantic vector expression represents semantic characteristics of the paired natural language query and the natural language candidate answers;
generating a textual vector expression based on a pair of the natural language query and the natural language candidate answer using a multi-layered neural network as textual analyzer, wherein the textual vector expression represents textual characteristics of the pair of the natural language query and the natural language candidate answer, wherein the textual characteristics comprise a degree of matching between the pair based on appearance of words;
generating a relevance score based on the combined semantic vector expression and the textual vector expression using a relevance score generator;
and
providing the relevance score for ranking the natural language candidate answer to the natural language query.

2 more elements

Dependencies are inferred from the claim text.