An article kansei retrieval system combining recommendation function and interaction design

In most article retrieval systems using Kansei words there exists a gap between user's Kansei and the system's Kansei model. Therefore, it is not always easy to retrieve the desirable articles. The purpose of this paper is to bridge this gap not to put a strain on users by combining the recommendation function and interaction design with four features. First, users can retrieve intuitively as the system visualizes retrieval space consisting of a torus type SOM (Self Organizing Maps). Second, users can find the most desirable article in any case by elimination methods to delete undesirable articles pointed by the user. Third, neural networks in the system learn user's Kansei based on the most desirable article to improve the retrieval accuracy. Fourth, users can search articles by arbitrary Kansei words, and can edit retrieval criteria as they please. In the evaluation experiments, the authors took actual paintings as the articles, and evaluated usability (effectiveness, efficiency and satisfaction), novelty and serendipity. These results were led by the synergetic effects of the recommendation function and interaction design.