An Interactive Environment for Kansei Data Mining

Nadia Bianchi-Berthouze

Abstract: Kansei engineering is a relatively recent discipline aimed at understanding and modelling (1) how the user's brain process subjective information and (2) how this information can be manipulated by a computer. In this paper we address the modelling of visual impression from the point of view of multimedia data mining. Visual impressions are impressions experienced when observing images. They are highly subjective, complex and difficult to explicit. We propose a methodological approach that takes into account the large amount of information involved in the mapping between images and visual impressions arising in an observer and eventually the way the observer expresses such impressions. From a computational point of view, the modelling process integrates different techniques of multimedia data mining to learn associations between image characteristics and impression words. The user assumes an active role in directing the system's mining activity through mechanisms of externalisation. The externalisation process is supported by a conceptual space endowed with tools that allow the user to express his/her mental process and naive models into formal specification. A WEB based meta-search engine to retrieve images by impression words has been developed. It is used as a support in the close loop of creating and testing new modelling hypothesis.