| CPC G06N 5/04 (2013.01) [G06F 16/334 (2019.01); G06N 20/00 (2019.01)] | 16 Claims |
|
1. A computer-implemented method, comprising:
identifying a natural language query;
translating the natural language query into an intermediate representation;
converting the intermediate representation into one or more query triples having two levels of information; and
performing, for each of the one or more query triples, relation linking between the respective query triple and a plurality of knowledge base triples;
wherein the intermediate representation includes a graph, where: nodes in the graphs represents concepts, and edges within the graph represent relationships between concept nodes;
wherein performing the relation linking includes inputting the one or more query triples into a machine learning environment;
wherein the machine learning environment outputs, for the one or more query triples, a plurality of relationships of the respective query triple with knowledge base triples, as well as scores indicative of a strength of similarity for the respective relationship;
training the machine learning environment using a distant supervision dataset, including:
identifying a knowledge base, the knowledge base including a plurality of natural language sentences and a plurality of knowledge base triples,
for individual ones of the plurality of knowledge base triples, selecting one of the plurality of natural language sentences within the knowledge base as a corresponding natural language sentence for the knowledge base triple, and
training the machine learning environment, utilizing the plurality of knowledge base triples and their corresponding natural language sentences;
determining an answer to the natural language query using the scores and the trained machine learning environment; and
outputting the answer for displaying on a display.
|