面向小样本命名实体识别的实体语义优先提示学习方法  

Entity semantic-priority prompt learning method for few-shot named entity recognition

作　　者：何丽[1,2] 曾骁勇 刘杰 段建勇[1,2] 王昊 He Li;Zeng Xiaoyong;Liu Jie;Duan Jianyong;Wang Hao(School of Information,North China University of Technology,Beijing 100144,China;CNONIX National Standard Application&Promotion Laboratory,Beijing 100144,China)

机构地区：[1]北方工业大学信息学院,北京100144 [2]CNONIX国家标准应用与推广实验室,北京100144

出　　处：《计算机应用研究》2024年第12期3622-3627,共6页Application Research of Computers

基　　金：新一代人工智能国家科技重大专项资助项目(2020AAA0109703);国家自然科学基金资助项目(62076167,U23B2029)。

摘　　要：近些年来,提示学习在小样本命名实体识别任务上得到了广泛应用。然而,命名实体识别仍然是一个令牌级的标记任务,在通过提示模板调动预训练知识时,很容易忽略新实体类型的语义信息。为此,提出了一个语义优先的提示学习方法。具体来说,首先检测少量示例中实体类型蕴涵的语义信息,然后将实体语义信息和询问实体位置的提示模板输入模型中,利用模型中的非自回归解码器并行预测来提取实体;此外,为了确保语义信息与实体类型的关联性,使用对比学习的方法来训练一个分类器,以去除与实体类型无关的语义信息;最后在两个常用的公共基准数据集上评估了所提方法,实验结果证明了该方法的有效性。In recent years,prompt learning has been widely applied in few-shot named entity recognition tasks.However,named entity recognition remains a token-level tagging task that can easily overlook the semantic information of new entity types when mobilizing pre-trained knowledge via prompt templates.To address this issue,this paper proposed a semantic-priority prompt learning method.Specifically,the method firstly detected semantic information implied by entity types in a few examples,then the entity semantic information,along with a prompt template for querying entity positions,was input into the model.The model utilized a non-autoregressive decoder for parallel entity extraction.Additionally,considering the correlation between semantic information and entity types,this paper employed a contrastive learning approach to train a classifier,aiming to remove semantic information unrelated to entity types.Finally,the proposed method was evaluated on two common public benchmark datasets,and the experimental results demonstrate its effectiveness.

关 键 词：小样本命名实体识别 提示学习 语义信息 对比学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]