Multi-view Attention Mechanism Learning for POI Recommendation

Abstract POI(point of interest) recommendation is a very necessary research field in both academic and commerce, however, predicting users’ potential points interest always faced with the problems data sparsity context semantics. Some studies have shown that graph embedding technology alleviates problem to certain extent. However, neither techniques nor unsupervised learning models can adaptively learn different effects multiple relations between users POIs, respectively. In view this, we leverage contextual information POIs build multi-view affinity graphs(e.g. User-User, POI-POI User-POI), latent representations based on Graph Embedding Attention mechanism, namely GEA model. particular, first construct graphs by using user’s social relationship, geographical distance check-in behaviour, embed them into low dimensional shared space representation POIs. Afterwards, order take advantage relationships final task, exploit attention mechanism obtain fused make according preferences. Finally, design multi-task objective function for joint optimization more accurate results. Extensive experiments Gowalla verified effectiveness our