大模型知识增强的教育年鉴实体识别  

Large model knowledge enhancement for entity recognition of education yearbook

作　　者：杨健鑫 吕学强[1] 游新冬 YANG Jianxin;LV Xueqiang;YOU Xindong(Beijing Key Laboratory of Internet Culture and Digital Dissemination Research,Beijing Information Science&Technology University,Beijing 100192,China)

机构地区：[1]北京信息科技大学网络文化与数字传播北京市重点实验室,北京100192

出　　处：《北京信息科技大学学报(自然科学版)》2025年第1期30-39,共10页Journal of Beijing Information Science and Technology University(Science and Technology Edition)

基　　金：国家自然科学基金项目(62171043,62202061);北京市自然科学基金项目(4232025);青海省创新平台建设专项(2022-ZJ-T02);北京市教委科研计划科技一般项目(KM202311232003,KM202311232002)。

摘　　要：针对传统命名实体识别方法和大语言模型在教育年鉴领域实体识别中的局限性,尤其是在术语复杂和实体边界模糊的情境下,提出了一种将大语言模型关键实体识别与传统特征提取相结合的教育年鉴实体识别方法。将大语言模型抽取的特征与增强的依存关系特征、位置信息特征相结合,提升实体识别的准确性和鲁棒性;通过层次注意力机制优化特征融合,在教育领域新闻数据集上对方法性能进行了评估。实验结果表明,该方法在处理教育年鉴文本中具有复杂背景信息和多样化实体类别的任务时,表现优于近期主流模型和各类大语言模型。To address the limitations of traditional named entity recognition(NER)methods and large language models in the field of entity recognition for education yearbook,especially in situations that involve complex terminology and ambiguous entity boundaries,an entity recognition method for education yearbook integrating key entity recognition by large language models with traditional feature extraction was proposed.The features derived from large language models were combined with enhanced dependency relation features and position information features,resulting in improved accuracy and robustness of entity recognition.Hierarchical attention mechanism was employed to optimize feature fusion,and the performance of the proposed method was evaluated on a news dataset in the education domain.Experimental results demonstrate that the proposed method outperforms recent mainstream models and various large language models in handling tasks involving complex contextual information and diverse entity categories within education yearbook texts.

关 键 词：命名实体识别 层次注意力机制 大语言模型 知识增强 教育年鉴 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]