NON-FACTOID TYPE QUESTION ANSWERING MACHINE

摘要

To provide a non-factoid type question answering machine capable of giving precise answers with high accuracy to non-factoid type questions.SOLUTION: A question answering device 124 comprises: a general word vector converting unit 210 for converting questions and answers into semantic vectors according to general context; general sentence level CNN 214 for weighting each semantic vector to calculate an expression at a sentence level of the question and answer corresponding to the similarity of the semantic vector between words in the question and words in the answer and the strength of causal relation between the words; a general passage level CNN 218 for weighing the sentence level expression in response to the similarity between the sentence level expressions of the question and the sentence level expressions of the answer and the strength of relation of the vector in the sentence level expression viewed from the causal relation, to calculate passage level expressions for the question and answer passage; and a classifier 184 for determining whether or not the answer is correct based on the similarity between the outputs of the CNN 214 and CNN 218.SELECTED DRAWING: Figure 10