Deep Weighted Graph Embedding for Link Weight Prediction

Graph structure is a widely existing data structure in real life. It has good modeling ability. Social networks, physical particles, texts, and so on, all can be represented as graph structure. Link prediction is a classical task in graph mining. This task uses the information of the observed network to predict missing links or possible new links between node pairs. Most of the existing studies are based on unweighted graphs. But in fact, lots of scenarios should be abstracted as weighted graphs, and the weight of links can reflect the strength of the ties between nodes. In this paper, we propose a deep weighted graph embedding method based on autoencoder and contrastive learning for link weight prediction on weighted graphs. Experiments on five real-world graph datasets demonstrate the effectiveness of our method.