Group Dynamics in Group Recommendation

Abstract

In this project we describe a model for group recommendation incorporating topic modeling techniques and influence scoring operations for improved accuracy in recommendation. Our model is motivated through the need to incorporate factors of group dynamics when recommending items to groups. We present our model for estimating user profiles, group profiles, and group member influence scores for group recommendation. Then, the precision of our proposed model is evaluated on three di↵erent types of groups artificially created from the Wikipedia dumps dataset against an implementation of a collaborative filtering recommender system with a least-misery aggregation heuristic. The successes of our model in this baseline test provide the basis for further study. We conclude by discussing limitations, further testing, and future research.