Neural-Symbolic Relational Reasoning on Graph Models: Effective Link Inference and Computation from Knowledge Bases

The recent developments and growing interest in neural-symbolic models has\nshown that hybrid approaches can offer richer models for Artificial\nIntelligence. The integration of effective relational learning and reasoning\nmethods is one of the key challenges in this direction, as neural learning and\nsymbolic reasoning offer complementary characteristics that can benefit the\ndevelopment of AI systems. Relational labelling or link prediction on knowledge\ngraphs has become one of the main problems in deep learning-based natural\nlanguage processing research. Moreover, other fields which make use of\nneural-symbolic techniques may also benefit from such research endeavours.\nThere have been several efforts towards the identification of missing facts\nfrom existing ones in knowledge graphs. Two lines of research try and predict\nknowledge relations between two entities by considering all known facts\nconnecting them or several paths of facts connecting them. We propose a\nneural-symbolic graph neural network which applies learning over all the paths\nby feeding the model with the embedding of the minimal subset of the knowledge\ngraph containing such paths. By learning to produce representations for\nentities and facts corresponding to word embeddings, we show how the model can\nbe trained end-to-end to decode these representations and infer relations\nbetween entities in a multitask approach. Our contribution is two-fold: a\nneural-symbolic methodology leverages the resolution of relational inference in\nlarge graphs, and we also demonstrate that such neural-symbolic model is shown\nmore effective than path-based approaches\n