Knowledge graph based question-answering model with subgraph retrieval optimization

Knowledge graph-based question answering (QA) is a critical domain within natural language processing, aimed at delivering precise and efficient responses to user queries. Current research predominantly focuses on minimizing subgraph sizes to enhance the efficiency and compactness of the search space. However, natural language queries often exhibit ambiguities, and merely reducing subgraph sizes may overlook relevant answer entities. Additionally, redundant relationships among entities in the knowledge graph can adversely affect QA model performance. To address these limitations, this paper introduces a novel QA model that optimizes subgraph retrieval. The proposed model enhances entity linking and subgraph retrieval by leveraging contextual features from both questions and entities. It disambiguates entities using relevant contextual features and refines the search process through entity relation merging and entity ranking strategies. This methodology improves entity recognition and linking, reduces subgraph dimensions, and broadens answer coverage, resulting in substantial improvements in QA performance. Experimental results on the CCKS2019CKBQA dataset demonstrate the model & sacute; effectiveness, showing an average F1 score improvement of 2.99% over the leading baseline model. Furthermore, the model's application in the field of ocean engineering underscores its practical utility and significance.