Novel Approaches to Group Recommendation

Abstract

Recommendation systems are extensively used to provide a constantly increasing variety of services. Alongside single-user recommendation systems, group recommendation systems have emerged as a method of identifying the items that a set of users will most appreciate collectively. In this thesis, we describe developments in the area of group recommendation techniques and how such techniques can be applied to address current challenges in the field of group recommendations. First, we propose a conceptual data model to support group recommendation that can be deployed in addition to the suggested approaches. Second, we propose a contextual group recommendation model that addresses the problem of contextual recommendation for groups and exploits a hierarchical context model to extend a typical recommendation model to a general context-aware model that addresses the information needs of a group. We also develop a context-aware recommendation system for concerts as a prototype for an exploratory analysis of the suggested model. Third, we propose a new dimension in the computation of group recommendations, namely, the exploitation of social ties (affinities) between group members, and its evolution over time; moreover, we present an efficient algorithm that produces temporal-affinityaware recommendations for ad hoc groups. Finally, we propose an approach that addresses the sparsity problem in group recommendation, and we present our method, which employs a memory-based technique to resolve the data sparsity problem in the group recommendation setting. All proposed methods have been evaluated through extensive experiments on public datasets or on real users who have participated in our experiments. Where possible, comparisons with related techniques have been performed to reinforce the validity of the presented approaches. Based on state-of-the-art metrics, the proposed methods have produced promising results for use in the field of group recommendation systems.