KDDLSA:A Knowledge Extraction Framework Integrating Deep Learning and Syntactic Analysis

Knowledge Extraction (KE) aims to extract entity and relationship information from text, serving as a foundational resource for downstream Natural Language Processing (NLP) tasks. While knowledge extraction methods in general domains are well-established, their performance diminishes when applied to specialized fields, as these domains often encompass unique professional information absent in broader contexts. In this paper, we introduce a knowledge extraction framework that amalgamates deep learning and syntactic analysis. This framework encompasses deep learning-based named entity recognition and relation extraction grounded in dependency parsing(KDDLSA). Specifically, we construct an external knowledge base tailored to specific fields through named entity recognition, thereby providing expert knowledge for relation extraction. Moreover, we devise a relation extraction algorithm by scrutinizing the nuances of Chinese grammar. Through the formulation of extraction rules, we pinpoint segments of the dependency syntax tree that adhere to these rules, enhancing the efficacy of knowledge extraction within specific domains. To validate our approach, we constructed a dataset in the maritime security domain using publicly available military news. Experimental results reveal a noteworthy performance enhancement when applying our method to the constructed dataset.