OPHAencoder: An unsupervised approach to identify groups in group recommendations

Recommender systems recommend items to users that would suit the users’ preferences. Suggesting personalized items in the context of a group of users is a non-trivial task. The increasing popularity of group recommender systems in recent years attracted researchers to compute the consensus among the group members more accurately. A recommendation is possible by aggregating the user preferences of the group. The composition of a group plays a significant role in group recommendation. As grouping is an unsupervised task, it becomes essential to form groups from the available information where each group member shares some common characteristics. In this paper, we have blended one permutation hashing and autoencoder techniques to auto-detect the groups. We use both methods very effectively to form the groups. We establish the efficacy of the proposed model in the order and flexible size preference models. We conducted experiments on real-world datasets and found that the proposed method is an efficient and robust approach to form a group automatically.