News-BERT: A Lexicon-Integrated Pre-Trained Framework for Enhanced Named Entity Recognition in the News Domain
Xiaojie Lin, J. L. Ping, Lei Cui, Helu Shan
Yantai Vocational College University of Science and Technology Beijing
阅读操作
确认中在文库中上传 PDF 后可生成中文音频讲解。
摘要与影响
Doing a good job of NER in news domain is helpful to build knowledge graph, realize information retrieval and clustering, and implement related news recommendation. In addition, it is also of great significance for public opinion analysis, brand management and decision-making of government, enterprises and media institutions. The existing named entity recognition methods are relatively lack of research in the news field. The traditional pre-trained models and algorithms have good performance on the recognition of commonly used named entities such as names, places, and times, but have a large gap in the recognition of entities in some proprietary fields. To this end, this paper proposes a News knowledge base enhanced bidirectional encoder representation method based on transformers, called News-BERT. Different from standard BERT+lexicon+CRF structures, News-BERT adopts parallel feature fusion of a news-specialized lexicon and Transformer encoding at character level, enabling better identification of long, nested, and domain-specific named entities, thus significantly improving recognition accuracy and domain applicability. We extensively experiment on the CLUENER2020 corpus and the People's Daily News NER corpus, two representative Chinese news datasets. Results show that our model achieves 79.20% precision and 78.61% F1 on CLUENER2020, and also outperforms baseline models on the People's Daily News NER dataset, demonstrating its strong generalization ability in the news domain.
逐年被引趋势
暂无年度引用数据
关键指标
同类平均 = 1
同领域 · 同年份 · 同类型
Google Scholar 与 OpenAlex 的被引统计范围不同,数值存在差异属正常。
AI 辅助阅读
依据:摘要
可就本文提问;依据不足时会说明。
学术脉络
学科主题
计算机 / AITopic Modeling
Natural Language Processing Techniques · Handwritten Text Recognition Techniques
参考文献 15
此处列出前 3 条