A Framework of Recommender System Using Interactive Evolutionary Computation

概要

論文の詳細を見る
This paper proposes the recommender system to retrieve multimedia data. Recommender systems using the collaborative filtering don't learn user's sensibility. In contrast, the system proposed us can learn user's sensibility. The system uses Kansei retrieval agent using the co-evaluation model and the only-evaluation model. The agent has a Kansei model controlled by Kansei parameters. These models are proposed in previous study, and are provided enough evidence of effectiveness. However, the previous study has an issue, how the agent is coded. Therefore, we propose a method of NPD cording. In this method, the agent has some elements. Each element has N (Negative) or P (Positive) or D (Don't care). In addition, this paper provides evidence of effectiveness the system and optimization performance of Kansei retrieval agent in simulations. In this simulation, we replace a real user with a simulant user. The simulant user is coded in the same way as an agent. We examined error between cords of the Kansei retrieval agent and cords of the mutual user. As a result, we confirmed what the error attenuates that the mutual user evaluate the presented data. In consequence, this paper confirmed that the system could present the data preferred the user.