KST-GCN: A Knowledge-Driven Spatial-Temporal Graph Convolutional Network for Traffic Forecasting

While considering the spatial and temporal features of traffic, capturing the\nimpacts of various external factors on travel is an essential step towards\nachieving accurate traffic forecasting. However, existing studies seldom\nconsider external factors or neglect the effect of the complex correlations\namong external factors on traffic. Intuitively, knowledge graphs can naturally\ndescribe these correlations. Since knowledge graphs and traffic networks are\nessentially heterogeneous networks, it is challenging to integrate the\ninformation in both networks. On this background, this study presents a\nknowledge representation-driven traffic forecasting method based on\nspatial-temporal graph convolutional networks. We first construct a knowledge\ngraph for traffic forecasting and derive knowledge representations by a\nknowledge representation learning method named KR-EAR. Then, we propose the\nKnowledge Fusion Cell (KF-Cell) to combine the knowledge and traffic features\nas the input of a spatial-temporal graph convolutional backbone network.\nExperimental results on the real-world dataset show that our strategy enhances\nthe forecasting performances of backbones at various prediction horizons. The\nablation and perturbation analysis further verify the effectiveness and\nrobustness of the proposed method. To the best of our knowledge, this is the\nfirst study that constructs and utilizes a knowledge graph to facilitate\ntraffic forecasting; it also offers a promising direction to integrate external\ninformation and spatial-temporal information for traffic forecasting. The\nsource code is available at\nhttps://github.com/lehaifeng/T-GCN/tree/master/KST-GCN.\n