NON-FACTOID QUESTION ANSWERING DEVICE

WIPO Patent Application WO/2019/012908

A1

[Problem] To provide a non-factoid question answering device capable of providing an appropriate, accurate answer to non-factoid questions. [Solution] A question answering device 124 includes: a general word vector conversion unit 210 for converting a question and an answer into a semantic vector in accordance with a general context; a general sentence level CNN 214 for calculating a sentence-level expression for the question and the answer by weighting each semantic vector in accordance with the degree of semantic-vector similarity between words in the question and words in the answer and the strength of cause-and-effect relationships between the words; a general passage level CNN 218 for calculating a passage-level expression for a question and answer passage by weighting a sentence-level expression in accordance with the degree of similarity with regard to question and answer sentence level expressions and the strength of relationships between vectors in a sentence-level expression from the perspective of a cause-and-effect relationship; and a classification unit 184 for determining whether an answer is a correct answer on the basis of the degree of similarity between the outputs from the CNNs 214 and 218.

OH JONGHOON (JP)
TORISAWA KENTARO (JP)
KRUENGKRAI CANASAI (JP)
IIDA RYU (JP)
KLOETZER JULIEN (JP)

PCT/JP2018/022696

January 17, 2019

June 14, 2018

Click for automatic bibliography generation  

NAT INST INF & COMM TECH (JP)

G06F17/30

OH, JONG-HOON ET AL.: "Multi-Column Convolutional Neural Networks with Causality-Attention for Why-Question Answering", PROCEEDINGS OF THE TENTH ACM INTERNATIONAL CONFERENCE ON WEB SEARCH AND DATA MINING, 2 February 2017 (2017-02-02), pages 415 - 424, XP058316669, DOI: 10.1145/3018661.3018737

SHIMIZU, Satoshi (JP)

View/Download PDF      PDF Help

Previous Patent: COMPUTATION DEVICE

Next Patent: SENSOR AND SENSOR-PRODUCING METHOD