近年の計算機の高速化、大容量化と低価格化、さらにはインターネットの普及によって情報化が加速度的に進んでいる。今後も増加しつづけるであろう膨大な情報や知識を適切に利用するためには情報の内容に関する、より高度な処理機能が要求される。このような機能を実現するためには、情報や知識の意味を理解しなければならない。そして、そのためには意味関係が表現できる構造化およびその利用のためのアプリケーションあるいはシステムが必要である。このような要求に対して、専門用語を最小単位とした概念構造の生成とその利用のためのシステムの開発を進めている。本研究では、概念構造を生成するのためのシステムの一部である階層関係の自動抽出法における問題点の調査について報告する。

抄録全体を表示

　The global flow of information is being developed at unprecedented speed. However, users can not sufficiently utilize huge amount of information by using conventional systems whose major functions are numerical calculation, symbol matching in information retrieval and deduction.Therefore, advanced utilization of contents of information are required. In order to realize such sophisticated utilization, it is necessary to understand meaning and characteristics of information. Therefore, the structuralization is required to represent various semantic relationships among information. In order to satisfy such requirement, we proposed a new representation of such structure and made a system for self organized knowledge resources based on semantic relationships.
　However, this system is a prototype and can not make enough structuralized knowledge resources to realize sophisticated utilization. Semantic relationships among knowledge resources must be correct and appropriate to objectives of applications. The main reason is that advanced utilization consists of navigations based on semantic relationships. Incorrect or inappropriate relationships influence such navigations. This paper reports problems which are classified and systematized at the method of an automatic extraction of hierarchical relationships which called SS-KWEIC because hierarchical relationships are the basis of semantic relationships in our concepts.

抄録全体を表示

We have developed a new flow-motion generation method that links the scaling and translation parameters of multi-view images with the position of a moving object in those images. Instead of calibrating extrinsic parameters, we used homographies between flat floors in images taken by each cameras, vanishing points in the vertical and horizontal directions, and the size of a reference object which was shot beforehand. We combined our new system with a system that tracks the positions of moving objects from multi-view images and demonstrated the flow-motion of moving objects.

抄録全体を表示