Query and Neighbor-Aware Reasoning Based Multi-hop Question Answering over Knowledge Graph

Multi-hop Question Answering over Knowledge Graph (multi-hop KGQA) is a challenging task since it requires reasoning with multiple triplets over knowledge graph to find the correct answer entities. Benefiting from expeditious development of attention mechanism and graph neural network, recent works in multi-hop KGQA based on information retrieval have made great progress. However, most existing works focus on encoding questions and knowledge graph in isolation so that the reasoning module lacks interaction of encoding information. Additionally, they only consider matching relation embeddings with the encoded queries at each hop, hence the complex questions, such as those containing one-to-many relation, are hard to answer. In this paper, we propose a query and neighbor-aware reasoning based multi-hop KGQA model to solve the above problems by introducing the CoAttention module and the neighbor-aware reasoning module. Experiments show that our model can not only keep competitive performance on MetaQA datasets but also improve performance than the state-of-the-art baselines on the wide-used benchmarks WebQSP and CWQ.