An Interpretable Question Answering Method Based on Heterogeneous Graph Neural Networks

Community question answering has gradually become the primary path for people to acquire knowledge. Existing community answer prediction research mainly faces the challenges of multi-type entity feature fusion and low answer interpretability. In this paper, we propose an interpretable Question Answering Method Based on Heterogeneous Graph Neural Networks, called IQAGNN. Firstly, we construct a heterogeneous information network to fuse multi-type entity relationships in question answering communities. Secondly, we employ the heterogeneous graph neural network to obtain multi-type entity embeddings. Finally, we predict the answer to the new question based on entity correlation and give an interpretation analysis of the answer via meta-paths. Comparative experiments on three real datasets show that IQAGNN has advantages in interpretive question-answering research.