POI location-aware services package recommendation for random groups

Recently, with the rapid growth of Point of Interest (POIs) and increasing of social activities, more and more individuals tend to join parties or trips with groups formed randomly (named random groups, RGs). Although some works have proposed wonderful POI recommendation methods for RGs, they failed to consider how to recommend service packages for RGs after they accept the recommended POIs, which significantly degrades user experience. To address this issue, this work proposes a POI location-aware service package recommendation model for random groups (termed PLSPR-RG). First of all, PLSPR-RG generates fitted feature representations of random groups with an enhanced feature learning method. Then, it selects users with similar characteristics to the random group and utilizes service usage data of similar users as the training dataset for PLSPR-RG. Subsequently, PLSPR-RG constructs a service labeled bipartite graph of similar users, then it employs an edge-enhanced bipartite graph neural network with multi-view contrastive learning (EBGNN-CL) to learn service feature representations that capturing both positive and negative interaction preferences of similar users. Furthermore, PLSPR-RG builds a directed service transition graph based on the service interaction records of similar users and adopts the SRGNN network to learn service feature representations containing latent user preferences, as well as user preference representations. Finally, based on the learned service representations and the fitted features of random groups, PLSPR-RG generates a ranked list of recommended services, thereby completing POI location-aware service package recommendation for random groups. Extensive experiments have been conducted on three public datasets (Foursquare, Gowalla and Yelp dataset), which have proved that our proposed model outperforms the state-of-the-art models.