| 摘要 |
PROBLEM TO BE SOLVED: To exactly estimate a non-determined word in a sentence while considering a context without requiring a large scale corpus for each topic. SOLUTION: The group of words to become candidates concerning a word, which is an estimation object in a sentence in question, is extracted and a word vector value knowledge based with a vector value added thereto, with which a distance between words is shortened as semantic similarity between the words become close, is retrieved. Thus, sentences up to just before the sentence in question are clustered and on the basis of the distribution of the sets of independent word vectors, sentence vectors or known independent word vectors in these several clusters, the right answer is found out. |
