Journal Article 音声ドキュメントを検索対象とした用語検索
Automatic Term Retrieval from Spoken Document

森田, 直樹  ,  南條, 浩輝  ,  山本, 凌紀  ,  馬, 青

58 ( 3 )  , pp.762 - 767 , 2017-03-15
Term retrieval from spoken document for video search is addressed. We assume that target terms likely occur at some parts of spoken document semantically similar to a given query, and try to find them by selecting passages and extracting related terms based on that assumption. As for passage unit, we propose an automatically divided unit which is just segmented according to the number of utterances. We evaluated the proposed method with 2,702 lecture speeches from Corpus of Spontaneous Japanese, and confirmed that 80% or more terms are found at an output candidate list, and almost 40% terms are listed within top 20.

Other information