融合偏旁特征的化学命名实体识别研究  

Research on Chemical Named Entity Recognition Combining Radical Features

作　　者：尹止戈 陆建峰[1] YIN Zhige;LU Jianfeng(Nanjing University of Science&Technology,Nanjing 210094)

机构地区：[1]南京理工大学,南京210094

出　　处：《计算机与数字工程》2023年第4期809-813,892,共6页Computer & Digital Engineering

基　　金：国家重点研发计划(编号:2018YFB1004904)资助。

摘　　要：化学实体识别是构建化学学科知识图谱,形成学科知识体系的基础步骤。中学化学课程语料多为非结构化文本数据,且具有语料少、相同词根相似语义的特性。传统的小规模特定领域的实体识别多基于规则和统计机器学习模型,需要大量的专家辅助和筛选,存在着可移植性弱,扩展维护难的问题。论文针对领域特性构建了基于BiLSTM-CRF的改进模型,融合了偏旁部首的词根向量特征进行化学实体识别。在人教版高中化学教材上进行对比实验,实验表明该方法在F1度量上高于实验中的基准模型,证明了此方法在中学化学知识点实体识别上的有效性。Chemical entity recognition is the basis step to construct chemical knowledge graph and form discipline knowledge system.The traditional entity recognition of small-scale domain-specific is mostly based on rules and statistical machine learning models,which requires a lot of expert assistance and screening,and has the problems of weak portability and difficult extension and maintenance.The high school chemistry curriculum corpus is mostly unstructured text data,and has the characteristics of low resources,the same root and similar semantics.According to the domain characteristics,this paper constructs an improved model based on BiLSTM-CRF,which combines radical embedding feature.In the PEP high school chemistry textbook,the experiment shows that this method is better than the benchmark model in the F1 measurement,which proves the effectiveness of this method in the chemistry knowledge points entity recognition.

关 键 词：深度学习 命名实体识别 偏旁特征 非结构化文本 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]