A Group Package Recommender Based on Learning Group Preferences, Multi-Criteria Decision Optimization, and Voting

This paper proposes a Group Package Recommender (GPR) framework, which provides recommendations on dynamically defined packages of products and services. It focuses on extending recommender systems in three ways: (1) to consider composite, rather than atomic, recommendations; (2) to deal with multiple, rather than single, criteria associated with recommendations; and, most importantly, (3) to support groups of users rather than individual users. This framework is based on: (1) defining the space of alternatives; (2) eliciting the utility function for each individual decision maker; (3) estimating the group utility function; (4) using the group utility function to find an optimal recommendation alternative; (5) constructing a set of diverse recommendations which contains the optimal recommendation alternative; and (6) applying alternative voting methods from social choice theories, to refine the recommendations. To evaluate the group recommender performance under each applied voting method, a preliminary experimental real-world user study is conducted, which shows that the proposed framework is able to produce a small set of recommendations that retains near optimal recommendations in term of precision and recall.