Topological graph representation learning on property graph

Property graph representation learning is using the property features from the graph to build the embeddings over the nodes and edges. There are many graph application tasks are using the property graph representation learning as part of the process. However, existing methods on Property graph representation learning ignore either the property features or the global topological structure information. We propose the TPGL, which utilizes the topological data analysis with a bias property graph representation learning strategy. The topological data analysis could augment the global topological information to the embedding and significantly improve the embedding performance on node classification experiments. Moreover, the designed bias strategy aggregated the property features into node embedding by using GNN. Particularly, the proposed TPGL outperformed the start of the art methods including PGE in node classification tasks on public datasets.