融合ERNIE和注意力机制的中文关系抽取模型  

Chinese Relation Extraction Model Based on ERNIE and Attention Mechanism

作　　者：李天昊 霍其润[1] 闫跃 徐远超[1] LI Tian-hao;HUO Qi-run;YAN Yue;XU Yuan-chao(College of Information Engineering,Capital Normal University,Beijing 100048,China)

机构地区：[1]首都师范大学信息工程学院,北京100048

出　　处：《小型微型计算机系统》2022年第6期1226-1231,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(62077002)资助;北京市自然科学基金项目(4212017)资助;首都师范大学交叉科学研究院资助.

摘　　要：关系抽取任务是要在实体识别的基础上确定无结构文本中实体对间的关系类别,即判断实体间的关系.针对目前中文关系抽取精度不足以及静态词向量无法很好地解读文本的问题,本文提出一种融合ERNIE预训练模型和注意力机制的TEXTCNN中文关系抽取模型.ERNIE词向量针对中文的特点以词组为单位做掩盖进行模型训练,实现了对中文文本更好的语义表达,再通过TEXTCNN模型对输入数据进行特征提取,融合注意力机制聚焦于影响最终结果的关键特征,从而实现特征优化提取.本文在百度发布的SKE数据集上进行实验,重点探索ERNIE模型结合注意力机制对中文文本的特征表达效果,结果表明本文模型可以更好地学习中文文本中的特征并用于关系抽取,有效提高关系抽取任务的准确率.Relationship extraction task is to determine the relationship category between entity pairs in unstructured text on the basis of entity recognition,thatis to judge the relationship between entities.In view of the problem thattheaccuracy of Chinese relation extraction is insufficient and static word vector can not interpret text well,this paper proposes a TEXTCNN Chinese relation extraction model based on ERNIE pre-training model and attention mechanism.ERNIE word vector is based on the characteristics of Chinese,and the phrase is used as the unit to cover up the model for training,so as to achieve better semantic expression of Chinese text.Then feature extraction is carried out on the input data through the TEXTCNN model,and the attention mechanism is integrated to focus on the key features that affect the final result,so as to achieve feature optimization extraction.This paper conducts experiments on the SKE dataset released by Baidu,focusing on the feature expression effect of ERNIE model combined with attention mechanism on Chinese text.The results show that the proposed model can better learn features in Chinese text and be used for relationship extraction,effectively improving the precision of relationship extraction task.

关 键 词：ERNIE TEXTCNN 注意力机制 关系抽取 中文文本 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]