Quantitative common sense estimation system and its application to automatic membership function generation

Systems capable of autonomous thinking and estimation have been required to cope with unknown situations. One of the important issues is knowledge, especially common sense, acquisition. This paper proposes new quantitative common sense estimation methods and applies them to an automatic membership function generation system. The proposed system estimates threshold values corresponding to large and small for various kinds of object-attribute sets to make membership functions. Here, the proposed system tries to relate each object and the impression. Two methods are proposed in this paper. The method-1 obtains data from top 1,000 snippets by Web search and estimates the global and local tendencies by clustering. The method-2 uses the number of hits in Web search together with parts of the results obtained by the method-1. In addition, several techniques are devised to eliminate unnecessary information from the retrieved Web pages. We carried out evaluation experiments: the effectiveness of the proposed methods has been shown and effectiveness of the combined method is indicated.