Abstract
There are case-based recommender systems that generate personalized recommendations for users exploiting the knowledge contained in past recommendation cases. These systems assume that the quality of a new recommendation depends on the quality of the recorded recommendation cases. In this paper, we present a case model exploited in a mobile critique-based recommender system that generates recommendations using the knowledge contained in previous recommendation cases. The proposed case model is capable of modeling evolving (conversational) recommendation sessions, capturing the recommendation context, supporting critique-based user-system conversations, and integrating both ephemeral and stable user preferences. In this paper, we evaluate the proposed case model through replaying real recommendation cases recorded in a previous live-user evaluation. We measure the impact of the various components of the case model on the system's recommendation performance. The experimental results show that the case components that model the user's contextual information, default preferences, and initial preferences, are the most important for mobile context-dependent recommendation.