Algorithm for Integrating Multi-Proximity for Trust-Based Group Recommendation in Ridesharing

It is anticipated that ridesharing will deliver solutions for transportation-related problems in large cities. A key issue of ridesharing is how to recommend a trustable ridesharing group. Because of the sparse rating on trust relations, large-scale group recommendation scalability is poor, which has led to difficulties in efficiently finding the most preferred carpoolers. Existing approaches usually study groups with a single type of geographic proximity or social proximity among participants. However, in reality, multiple proximities exist between users, yielding a combination and balance of different proximities for finding the groups. In this work, we formalize the problem of group recommendation via an attributed multiplex network, and we form the Trusted Ridesharing Group (TRG) that preserves ridesharing participants' trustworthiness. Our proposed algorithm takes advantage of strong representative power and models these proximities across different networks and then provides path-based explanations for group recommendations. Moreover, our solution benefits from an attention-based weighted collaboration, enabling the passengers to be grouped by their interests. Experimental results on a popular social dataset show the proposed method's superiority over state-of-the-art methods for network representation learning and group recommendation.