TrEKBQA:Traversing Knowledge Graph Embedding for Multi-hop Knowledge Base Question Answering

Abstract

Abstract Recent research apply KG embedding to multi-hop Knowledge Base Question Answering(KBQA) to predict missing links, however, it is often affected by the skewed distribution of nodes in the knowledge graph, resulting in poor generalization of the model. Therefore, we propose a method TrEKBQA based on traversing the knowledge graph embedding space for multi-hop KBQA, which performs path traversal in the KG embedding space instead of KG itself for link prediction to complete the knowledge graph, thus improving the accuracy of multi-hop KBQA.TrEKBQA model complex relationships using correlations between individual links and longer paths connecting the same pair of entities to traverse the KG embedding space to mitigate the effects of biased distribution of nodes and improve the performance of link prediction. In the pre-processing process, TrEKBQA uses the PRN algorithm to extract subgraphs related to the problem entity to reduce the number of target entities. Through experiments on multiple benchmark datasets, we demonstrate the effectiveness of TrEKBQA on KBQA tasks.