Terminology Extraction Based on Document-Level Context and Domain Adaptation

Automatic terminology extraction holds significant application value in natural language processing tasks, particularly in domain-specific knowledge graph construction, information retrieval, and machine translation. However, existing terminology extraction methods demonstrate limited performance in cross-domain scenarios due to substantial variations in linguistic conventions, term distributions, and contextual dependencies across different domain-specific texts. This paper proposes a novel terminology extraction model that integrates document-level contextual features with domain adaptation techniques. The model achieves effective alignment between source and target domain distributions through cross-sentence dependency analysis and reverse Kullback-Leibler divergence optimization, thereby enhancing generalization capabilities. Experimental results demonstrate that: (1) Document-level contextual features yield an average 3.74% improvement in F1-score; (2) The domain adaptation mechanism provides additional 2.32% F1-score enhancement, particularly improving model adaptability in low-resource target domains. Our methodology offers new perspectives for applying terminology extraction in multi-domain scenarios.