基于XLNet的情感分析模型  

Emotion Analysis Model Based on XLNet

作　　者：梁淑蓉 谢晓兰[1,2] 陈基漓 许可[1] LIANG Shu-rong;XIE Xiao-lan;CHEN Ji-li;XU Ke(College of Information Science and Engineering,Guilin University of Technology,Guilin 541004,China;Guangxi Key Laboratory of Embedded Technology and Intelligent Systems,Guilin 541004,China)

机构地区：[1]桂林理工大学信息科学与工程学院,桂林514004 [2]广西嵌入式技术与智能系统重点实验室,桂林514004

出　　处：《科学技术与工程》2021年第17期7200-7207,共8页Science Technology and Engineering

基　　金：国家自然科学基金(61762031);广西科技重大专项(桂科AA19046004);广西重点研发项目(桂科AB18126006)。

摘　　要：目前通过深度学习方法进行语言模型预训练是情感分析的主要方式,XLNet模型的提出解决了BERT模型上下游任务不一致的问题。在XLNet基础上增加LSTM网络层和Attention机制,提出XLNet-LSTM-Att情感分析优化模型,通过XLNet预训练模型获取包含上下文语义信息的特征向量,接着利用LSTM提取上下文相关特征,最后引入注意力机制根据特征重要程度赋予不同权重,再进行文本情感倾向性分析。仿真实验中将XLNet-LSTM-Att模型与5种常用的情感分析模型进行对比,结果表明提出的模型优于其他测试模型,模型的精准率达到89.29%。At present,deep learning method is the main way to pre-training language model.The XLNet model solves the problem of inconsistency between the upstream and downstream tasks of BERT model.LSTM network layer and Attention mechanism was added to XLNet,and XLNet-LSTM-Att sentiment analysis optimization model was proposed.The feature vectors containing context semantic information were obtained by XLnet pre-training model,and then the context related features were extracted by LSTM.Finally,the attention mechanism was introduced to give different weights according to the importance of the features,and then the text sentiment tendency was analyzed.In the simulation experiment,the XLNet-LSTM-Att model was compared with five commonly used sentiment analysis models.The results show that the proposed model is better than other test models,and the accuracy of the model reaches 89.29%.

关 键 词：情感分析 XLNet LSTM 注意力机制 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]