Point-of-interest prediction based on multimodal knowledge graph embedding

Point-of-interest prediction is one of the important functions of location-based information services and applications, nowadays there are research methods still have two problems, one is that data sparsity has a serious impact on the results of prediction, and the other is that it is incomplete for mining the implicit features in check-in data. In order to solve these problems, a point-of-interest prediction algorithm based on the fusion of semantic and structural features of knowledge graph is proposed. Firstly, the knowledge graph is constructed based on the check-in data from three perspectives of socialization, check-in record information, and point-of-interest information; secondly, from the structural and semantic perspectives of the knowledge graph, the R-GCN model and the word-vector model are introduced to obtain the embedded representations of the entities and relations in the knowledge graph weighted fusion, respectively; and finally, the vector representations of the entities and relations are used as inputs to the graph-attention network model for predicting the list of points-of-interest. The model has been subjected to several experiments on real datasets and compared with the baseline model, which proves that the model in this paper has better accuracy in the destination prediction problem.