基于ERNIE-BMBD-CRF模型的中文课程问句命名实体识别方法研究  

Research on Chinese course question named entity recognition method based on ERNIE-BMBD-CRF model

作　　者：秦慧伶 邢雪枫 张黔川 陈嘉豪 陈雅茜[1] QIN Huiling;XING Xuefeng;ZHANG Qianchuan;CHEN Jiahao;CHEN Yaxi(School of Computer and Artificial Intelligence,Southwest Minzu University,Chengdu 610041,China;School of Electronic and Information,Southwest Minzu University,Chengdu 610041,China)

机构地区：[1]西南民族大学计算机与人工智能学院,四川成都610041 [2]西南民族大学电子信息学院,四川成都610041

出　　处：《西南民族大学学报(自然科学版)》2025年第1期85-91,共7页Journal of Southwest Minzu University(Natural Science Edition)

基　　金：西南民族大学中央高校优秀学生培养工程项目(2023NYXXS041);西南民族大学横向项目(横20240096)。

摘　　要：随着智慧教育的快速普及,如何将课程知识点以更直观的形式进行展示,帮助学生更加高效地学习,已成为教育信息化领域一项重要的研究课题.传统中文命名实体识别方法因忽略文字上下文联系而不易推出被掩盖的文字,因此提出一种基于ERNIE-BMBD-CRF(EBC)模型的课程问句命名实体识别方法,以有效地从课程问句中识别出知识点实体.首先利用ERNIE预训练语言模型对文本词向量进行表征,然后构建BMBD层,通过BiLSTM模型进行上下文语义特征提取,将边界扩散机制引入到多头注意力机制MHA中以增强实体边界信息的捕捉.最后在CRF模型中进行序列标签解码,从而完成命名实体识别.实验结果表明:该模型在CLUENER2020、MSRA两个公开数据集和自建课程语料数据集上的F1值分别为87.45%、98.48%、81.49%,均优于其他对比模型。With the rapid popularization of smart education,how to present course knowledge points in a more intuitive form to help students learn more efficiently has become an important research topic in the field of educational informatization.Traditional Chinese named entity recognition methods are difficult to extract obscured text due to ignoring contextual connections.There-fore,a course question named entity recognition method based on ERNIE-BMBD-CRF(EBC)model was proposed to effectively identify knowledge point entities from course questions.Firstly,the ERNIE pre-trained language model was used to represent the text word vectors,and then the BMBD layer was constructed.The BiLSTM model was used to extract contextual semantic features,and the boundary diffusion mechanism was introduced into the multi head attention mechanism MHA to enhance the capture of entity boundary information.Finally,sequence label decoding was performed in the CRF model to complete named entity recognition.The experimental results showed that the Fl values of this model on two public datasets,CLUENER2020 and MSRA,as well as a self built course corpus dataset,were 87.45%,98.48%,and 81.49%,respectively,which were superior to other comparative models.

关 键 词：教育信息化 命名实体识别 ERNIE预训练语言模型 边界扩散 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]