Evaluating functional requirements-based compound critiquing on conversational recommender system

Conversational recommender system (CRS) is a form of recommender system that can refine user preference through conversational mechanism. User preference refinement can be proceeded in reference to the user's feedback towards the products recommended, called as critiquing technique. Compound critiquing technique has been widely developed to ensure the interaction efficiency in CRS. However, the compound critiques offered refer to the product technical features. In terms of hi-tech products, not all consumers are familiar with technical features. A model for generating functional requirement-based compound critiques (instead of technical features-based) has been developed in our previous work. In this paper, we evaluate this model from the aspects of recommendation accuracy, query refinement, and user satisfaction. The user study involving 88 users (either familiar or unfamiliar with technical features) shows that the approach has successfully increased the users' positive perception compared to the recommender system commonly used in e-commerce. Besides, this approach has a high recommendation accuracy (89.77%) and has successfully refined the users' needs.