A novel meta-graph-based attention model for event recommendation

Due to the popular trend of combining online and offline interactions among users in event-based social networks (EBSNs), event recommendation which helps users discover their interesting events has become progressively urgent. Different from classic item recommendation, candidate events usually have short life cycle and occur in the future, resulting in severe challenges of data sparsity and cold-start. However, these problems are not well studied by previous works. In this article, we propose a Meta-Graph-based Attention Recommendation (MGAR) model to tackle aforementioned challenges by fully exploring complex semantic information based on meta-graphs extracted from EBSNs. First, we model the interactions between different entities as a heterogeneous information network and construct multiple meta-graphs to characterize the latent semantic preferences of users. Subsequently, we utilize convolutional neural networks and attention mechanisms to learn user and event latent factors by extracting semantic features of meta-graphs. Furthermore, the fused latent features are utilized to predict the ratings of a user to events and the events with top-k scores are recommended to the user. We collect several real-world datasets from a popular EBSN platform and conduct extensive experiments on the datasets. Our proposed model attains superior recommendation performance over several state-of-the-art approaches. Moreover, the results demonstrate that meta-graphs can reveal the semantic properties between users and events and improve event recommendation performance.