KBDR: document retrieval based on graph matching with knowledge enhancement

Document retrieval refers to the identification of document contents that are semantically most relevant within a corpus. Existing matching algorithms may not recognize the synonymy of all documents for vocabulary, leading to retrieval inaccuracies. To address this limitation, this paper proposes Knowledge Based Document Retrieval (KBDR) that utilizes graph-based knowledge expansion. Firstly, named entity recognition and relation extraction are used to construct document graphs. Secondly, it expands entities using external knowledge bases such as HowNet and ConceptNet to enrich semantic connections. Thirdly, applies a Graph Attention Network (GAT) to model inter-entity dependencies for relevance ranking. Furthermore, we use five (i.e., CIPP, CNSE, CNSS, AAN, and OC) datasets to evaluate, KBDR outperforms baselines, achieving the highest Recall@1 of 25.88% on the CNSE. Overall, the KBDR can enhance efficiency and robustness in document retrieval, addressing conceptual deviations and advancing the field toward more complicated retrieval systems.