Affective Information Retrieval System

Sensitivity search in light of user preferences

Today, any desired information can readily be acquired by means of a keyword search. However, it can be difficult to find the information we really want because search engines often return such a huge number of items.

For example, buying fashion items online can be very hard work, as we must check all the recommended products one by one to identify one that matches our taste.

To solve precisely such a problem, we are working on a version of the Kansei search system informed by the taste of the user, with a view to developing a Kansei retrieval system for clothes and fashion goods.

Learning the user’s Kansei

The Kansei retrieval system is initially unable to understand user preferences and cannot therefore provide appropriate information. The system needs to learn the user’s Kansei to improve its retrieval ability.

So, how will computer understand user preferences? Let’s consider the issues involved in learning a user’s clothing preferences.

First, the system randomly selects some clothes from the clothing database and shows them to the user. The user evaluates these, assigning values to each item on the system. The system also evaluates these clothes, based on characteristics that affect user preference such as color, shape, pattern, and so on.

Based on differences between evaluation by self and user, the system then learns user preferences.

By repeating the learning process, the system can understand user preferences and major factors affecting user impressions.

Clothing retrieval system

We have developed a prototype system for clothing retrieval and evaluated the performance of the system. Experimental results suggest that the system can learn user preferences by use of the interactive genetic algorithm to identify clothes that each individual user will like.

However, the system has not yet acquired a complete understanding of user preferences, as individual sensitivity is very complicated.

The challenge here is to develop more effective methods for learning user Kansei and more useful sensitivity search systems that will be more convenient than conventional keyword-based systems.


Takaki Urai, Masataka Tokuamaru, “User Kansei Clothing Image Retrieval System”, Journal of Advanced Computational Intelligence and Intelligent Informatics, Vol.18, No.6 pp. 1044-1052, 2014-11.

T.Urai, D.Okunaka, M.Tokumaru, “Clothing image retrieval based on a similarity evaluation method for Kansei retrieval system”, Joint 6th International Conference on Soft Computing and Intelligent Systems and 13th International Symposium on Advanced Intelligent Systems (SCIS&ISIS2012), pp.261-265, 2012-11 (Kobe, Japan).