Machine Learning with Knowledge Graphs and Neural Networks

Graph data is being produced and collected at an accelerating pace in numerous state of the art ap- plications, and the structure and node properties of the graph data offer fertile ground for data mining and machine learning. Numerous machine learning models have already been applied to graph data and new models are being developed to specifically exploit it. In this work we examine distinct approaches for machine learning over graph data; we provide an in depth examination of applying a modern machine learning framework - probabilistic soft logic - to the problem of graph node label prediction and compare the results to novel neural network architectures applied to the same problem. We also examine the appli- cation of a novel knowledge graph based neural network architecture applied to the problem of vehicle traffic flow prediction and compare those results with well established neural network architectures for time series forecasting.