首页|期刊导航|青岛大学学报(自然科学版)|基于本体与知识增强的军事命名实体识别模型

基于本体与知识增强的军事命名实体识别模型OA

Military Named Entity Recognition Model Based on Ontology and Knowledge Enhancement

中文摘要英文摘要

针对军事领域命名实体识别中存在的知识稀疏与语义歧义问题,提出了一种融合本体与知识增强的命名实体识别模型(Ontology-Knowledge-enhanced Representation NER,OKR-NER),基于 BERT-BiLSTM-Attention-CRF框架,引入领域先验知识,通过双流编码、本体引导注意力机制(OGA)以及条件随机场(CRF)约束实现语义特征与领域知识的深度融合.研究结果表明,模型在自建军事语料上的F1值较BERT-CRF、SpanNER、Global-Pointer等基线模型提升2.41%~7.30%,消融实验验证了3项机制均有助于提升性能,其中 OGA贡献最显著;可有效缓解军事领域命名实体识别中的知识稀疏与语义歧义问题,提升军事实体识别效果.

To address persistent issues of knowledge sparsity and semantic ambiguity in military named entity recognition(NER),a domain-specialized ontology and knowledge enhanced model termed OKR-NER(Ontology-Knowledge-enhanced Representation for NER)was proposed.The model extends the BERT-BiLSTM-Attention-CRF architecture by incorporating structured domain knowledge,and achieves deep semantic-knowledge in-tegration through a dual-stream encoding strategy,an ontology-guided attention mecha-nism(OGA),and a knowledge-aware conditional random field(CRF)decoding module.The proposed model achieves F1-score gains of 2.41%~7.30%over strong baselines,in-cluding BERT-CRF,SpanNER,and GlobalPointer,on a self-constructed military corpus.All three proposed mechanisms contribute to performance improvements,with the ontolo-gy-guided attention(OGA)mechanism playing the most prominent role.The model effec-tively mitigates knowledge sparsity and semantic ambiguity in military named entity rec-ognition and improves military entity recognition performance.

王华伟;云红艳;张栋

青岛大学计算机科学技术学院,青岛 266071青岛大学计算机科学技术学院,青岛 266071青岛大学计算机科学技术学院,青岛 266071

信息技术与安全科学

命名实体识别军事领域深度学习本体增强双向编码器表示模型

named entity recognitionmilitary domaindeep learningontology enhance-mentbidirectional encoder representations from transformers

《青岛大学学报(自然科学版)》 2026 (3)

24-31,8

山东省高校教改项目重点项目(批准号:Z2023238)资助.

10.3969/j.issn.1006-1037.2026.03.05

评论