Item-driven group formation

Valcarce D., Brilhante I., Macedo J. A., Nardini F. M., Perego R., Renso C.

Group Recommendation  Recommender Systems  Group Formation  Information Systems  Location-Based Social Networks  Computer Networks and Communications  Communication 

Several daily activities, such as traveling to a tourist attraction or watching a movie in the cinema, are better enjoyed with a group of friends. However, choosing the best companions may be difficult: we need to consider either the relations among the chosen friends and their interest in the proposed destination/item. In this paper, we address this problem from the perspective of recommender systems: given a user, her social network, and a (recommended) item that is relevant to the user, our User-Item Group Formation (UI-GF) problem aims to find the best group of friends with whom to enjoy such item. This problem differs from traditional group recommendation and group formation tasks since it maximizes two orthogonal aspects: (i) the relevance of the recommended item for every member of the group, and (ii) the intra-group social relationships. We formalize the UI-GF problem and we propose two different approaches to address it. In the first approach, the problem is modeled as the densest k-subgraph problem over a specific instance of the social network of the user, while the second approach is based on a probabilistic collaborative filtering method that exploit relevance-based language models. We perform an extensive assessment of several algorithms solving the two approaches in the domain of location recommendations by exploiting five publicly available Location-Based Social Network (LBSN) datasets. The experimental results achieved confirm the effectiveness and the feasibility of the proposed solutions that outperform strong baselines. Indeed, results reveal interesting and orthogonal properties of the two formulations. The probabilistic collaborative filtering approach is more effective than the graph-based one on datasets with sparse social networks but with more dense check-in data. On the contrary, the graph-based model performs very well on datasets which present high sparsity on the ratings and check-ins but a higher number of links among users.

Source: Online social networks and media 8 (2018): 17–31. doi:10.1016/j.osnem.2018.10.002

Publisher: Elsevier, Amsterdam, Paesi Bassi

Metrics