An Efficient Method to Retrieve Passages in Documents Using Field-reminding Words

摘要

Passage Retrieval is the very, useful technique for various natural language processing systems, such as information retrieval systems, spoken dialogue systems and Kana-to-Kanji conversion systems. This paper proposes the method to decide the passages using the field-reminding words, which can be directly related to the field of a text. This method decides the range of the passage based on five association levels and positions of field-reminding words in the text, and the topic field of each passage is specified by the field name with which these words are associated. Moreover, the algorithm proposed in this paper can avoid the overlap between neighboring passages. We evaluate 120 Japanese documents with the presented method, and verify the usefulness with 88% for precision and 69% for recall. We propose a usefulness of the presented method after we can compare the passage of our approach into one of human judgment.