Retrieval-Augmented Knowledge Graph Reasoning for Commonsense Question Answering

Existing knowledge graph (KG) models for commonsense question answering present two challenges: (i) existing methods retrieve entities related to questions from the graph, which may extract noise and irrelevant nodes, (ii) there is a lack of interaction representation between entities. However, current mainly focus on retrieving relevant with some noisy nodes. In this paper, we propose novel retrieval-augmented (RAKG) model, solves above issues using key innovations. First, leverage density matrix make model reason along corrected path an enhanced subgraph graph. Second, fuse representations through bidirectional attention strategy, in update convolutional network (GCN). To evaluate performance our method, conducted experiments widely used benchmark datasets: CommonsenseQA OpenBookQA. The case study gives insight into finding that augmented provides reasoning answering.