合金研究論文テキストからの知識抽出

抄録

An information base system to support construction and exploration of an information space in metallurgy, especially in the domain of superalloys, was implemented. Information about materials development is, in most cases, obtained empirically. Use of natural language is, therefore, important to transmit such information. The developed system interprets technical papers on metallurgy and constructs information space semi-automatically. It consists of three sub-systems. The first retrieves information from incomplete keys by using thesaurus. Successful results were obtained by picking up all the idiomatic expressions appeared in the articles on the subject. The second sub-system, called METIS(MET allugy papers Intelligent Surveyors) adopts a naive natural language process. The heart of METIS is a packet of domain specific knowledge called KP (Knowledge Pieces) in which procedures for extracting and structuring technological information are embedded. METIS extracts technological information from technical papers on metallurgy written in a mark-up language. Products of METIS is a variety of summaries and surveys such as structured technical summary, as well as visualization of similarities, differences of relevant papers, and cause-effect relations. The third sub-system is called KE (Knowledge Editor) adopts both syntactic and semantic analysis technologies. In the system, domain knowledge is represented using an instance-oriented and script-based method. KE represents knowledge on metallurgy by a set of objects together with a set of relations between them. KE provides two interactive facilities for exploring into information space; one is the natural language query based on semantic understanding and the other is the navigation of the relations between objects.