Now, searching for what you want, there is a convenient thing called the Internet. When searching, enter “keyword” such as “bag” or “T shirt”. But here is a little problem.
Even saying “bag”, there are various kinds as rucksacks, tote bags and handbags. Also the design is various. Like a backpacks and tote bags, “what you can be expressed by words” can be narrowed down by keywords. However, it is difficult to express the design in words. Even if expressed in words like “cool bags” or “pretty bags”, there are individual differences such as “I think it is cool, other don’t think so much”.
The sensitivity of people is different. Modern search systems don’t fully understand human sensitivity and show search results.
In this kind of search system, we are conducting research related to basic model of system that can understand people’s sensitivity and can recommend products. So far, we have proposed and verified a model called Kansei retrieval agent. At first, Kansei retrieval agent learns the tendency of evaluation on user’s data. After that, the agent acts like a concierge to search what the user likes from a huge database instead of the user.
System outline of Kansei retrieval agent.
Various meta-heuristics techniques such as evolutionary computation and pattern recognition are applied to the Kansi retrieval agent.
We are researching a method to implement a sensibility search agent model using fuzzy inference. Fuzzy inference quantifies the physical characteristics of an object using ambiguous scales that are closer to human perception (such as small/large, cold/hot) through membership functions. It then performs fuzzy set operations using these membership values and multiple fuzzy rules to determine the inference value. This research has the advantage of potentially extracting users' latent preference rules. If we can identify preference rules that users themselves may not even be aware of, companies could utilize this information for future product development or recommend new design products to users.
In this research, we have built rule extraction systems based on character coordination and J-POP music, conducting various validations.
For more details, click here. → Character Coordination，Music Search
Recently, various inferences and predictions have been achieved through deep learning methods. However, it is often "difficult to explain the reasoning behind these inference results in a way that is understandable to humans." Fuzzy inference, a technology that has been around for a long time in the field of information technology, holds potential to help address these issues.
Related our works
[International Conference] Ryota Shiraishi, Hiroshi Takenouchi, Masataka Tokumaru, "Optimization of Fuzzy Rules in Kansei Retrieval Agent with Fuzzy Reasoning", Joint 10th International Conference on Soft Computing and Intelligent Systems and 19th International Symposium on Advanced Intelligent Systems (SCIS&ISIS2018), pp.449-454, 2018-12 (Toyama, Japan).
[Academic Papers] Hiroshi Takenouchi, Masataka Tokumaru, "Kansei Retrieval Agents Model with Fuzzy Reasoning", International Journal of Fuzzy Systems, Vol.19, Issue.6, pp.1803-1811, 2017-12.
*Bold: Students in this research laboratory
