Enhancing question answering in educational knowledge bases using question-aware graph convolutional network

In this paper, a question answering method is proposed for educational knowledge bases (KBQA) using a question-aware graph convolutional network (GCN). KBQA provides instant tutoring for learners, improving their learning interest and efficiency. However, most open domain KBQA methods model question sentences and candidate answer entities independently, limiting their effectiveness. The proposed method extracts description information and query entity sets for a specific question, processes them with Transformer and pre-trained embeddings of the knowledge base, and extracts a subgraph of candidate answer sets from the knowledge base. The node information is updated by GCN with two attention mechanisms expressed by the question description and query entity set, respectively. The query description information, query entity set, and candidate entity representation are fused to calculate the score and predict the answer. Experiments on MOOC Q&A dataset show that the proposed method outperforms benchmark models.