Grant-in-Aid for Scientific Research (B).
|Research Institution||Science University of Tokyo|
FUJISAKI Hiroya Science University of Tokyo, Faculty of Industrial Science and Technology, Professor, 基礎工学部, 教授 (80010776)
OHNO Sumio Tokyo University of Technology, School of Engineering, Assistant Professor, 基礎工学部, 講師 (80256677)
KAMEDA Hiroyuki Tokyo University of Technology, School of Engineering, Associate Professor, 工学部, 助教授 (00194994)
|Project Fiscal Year
1998 – 1999
Completed(Fiscal Year 1999)
|Budget Amount *help
¥9,000,000 (Direct Cost : ¥9,000,000)
Fiscal Year 1999 : ¥2,200,000 (Direct Cost : ¥2,200,000)
Fiscal Year 1998 : ¥3,400,000 (Direct Cost : ¥3,400,000)
Fiscal Year 1997 : ¥3,400,000 (Direct Cost : ¥3,400,000)
|Keywords||information network / intelligent system for information retrieval / key concept / spoken dialogue / unknown word / polysemy / synonymy / dialogue modeling / 情報ネットワーク / 知的情報検索システム / キー概念検索 / 音声対話 / 未知語 / 同表記異義 / 異表記動議 / 対話モデル / エージェント / 未知語処理 / 意図推察 / インターネット / キ-概念検索|
The project aimed at realizing an intelligent system for information retrieval over the Internet, The main results obtained are as follows.
Use of 'key concepts', estimation of relevance of retrieved results, and spoken dialogue as the user interface, were adopted as basic principles.
(2)Processing of unknown words
In order to conduct search based on key concepts, the system has to infer the concepts of key words that are not registered in the system's lexicon. On the basis of an extensive collection and classification of such 'unknown words', methods were developed for processing unknown words arising from variations of transcription as well as unknown compound words consisting of known morphemes.
(3)Processing of polysemy and synonymy
In order to realize concept-based search, both polysemy and synonymy of keywords have to be coped with. As for polysemy, a method for disambiguation was developed on the basis of collocation information. As for synonymy, a method was developed to expand a keyword on the basis of its concept.
(4)Relevance estimation of retrieved results
A method was developed for estimating the degree of relevance of a document to a query on the basis of number and location of occurrence of keywords within a document. The estimation was optimized to maximize the correlation between the estimated relevance score and the actual score based on human judgment.
(5)Dialogue management based on user and system modeling
A method for dialogue management was developed on the basis of analysis of simulated dialogues between a user and the system. It adopts separate modeling of the user and the system, represented by two finite-state automata exchanging information mainly through their utterances.
(6)Construction of a prototype system
A prototype system was constructed combining the above-mentioned results, and its validity was tested and confirmed experimentally.