An article retrieval support system that learns user's Kansei

Abstract

Most of article retrieval systems using retrieval criteria of Kansei words have a gap between user's Kansei and system's Kansei model. Therefore, it is not always easy to retrieve the desired articles efficiently according to the user's preference. This paper proposed a system to retrieve the desired articles quickly and intuitively from the database. To achieve this aim, dimension of the retrieval space is compressed by a torus SOM (Self Organizing Maps), and a user can move in the retrieval space panoramically. A user can also choose an elimination method during search. By this method, the system estimates the significant Kansei parameters and makes the search more efficient. The system also has a function to eliminate the unselected articles and reduces the size of SOM. Additionally, the system learns the Kansei of individual user from the retrieval results by using neural networks. In evaluation experiments, we took actual painting as article, and confirmed the efficacy of the proposed method.