基于BE-MCNN模型的新闻评论情感分析方法  

News Commentary Sentiment Analysis Method Based on BE-MCNN Model

作　　者：李文书[1] 管平 LI Wenshu;GUAN Ping(College of Artificial Intelligence,Zhejiang Sci-Tech University;College of Information Science and Technology,Zhejiang Sci-Tech University,Hangzhou 310018,China)

机构地区：[1]浙江理工大学人工智能学院 [2]浙江理工大学信息学院,浙江杭州310018

出　　处：《软件导刊》2024年第3期1-7,共7页Software Guide

基　　金：国家自然科学基金项目(31771224,61603228);国家重点研发计划项目(2018YFB1004901);浙江省自然科学基金项目(LY17C090011,LGF19F020009)。

摘　　要：实时新闻评论具有文本短、信息丰富、结构复杂等特点,情感分析难以准确捕捉其真实的情感倾向。为增强语义的特征信息,减少模型过拟合问题,提高新闻评论情感分析的准确性,提出一种融合BERT模型、Transformer En⁃coder与多尺度CNN模型的新闻评论情感分析算法。首先,针对新闻评论长度较短、表达情绪观点内容较多的特点,使用BERT模型对新闻评论文本进行预训练,获得具有上下文信息的特征向量;其次,为解决模型过拟合问题,在BERT模型下游添加一层Transformer编码器;最后使用四通道双层CNN模型,通过组合不同大小尺寸的卷积核来提升模型分析新闻评论情感的性能。实验结果表明,该方法在两个新闻评论数据集上的准确率分别达到93.0%与96.4%;与不同模型的比较实验进一步证明了所提方法的有效性。Real time news comments have the characteristics of short text,rich information,and complex structure,making it difficult for sentiment analysis to accurately capture their true emotional tendencies.To enhance semantic feature information,reduce model overfitting problems,and improve the accuracy of news comment sentiment analysis,a news comment sentiment analysis algorithm is proposed that inte⁃grates BERT model,Transformer Encoder,and multi-scale CNN model.Firstly,in response to the short length of news comments and the high content of expressing emotional views,a BERT model is used to pre train news comment texts and obtain feature vectors with contextual information;Secondly,to solve the problem of model overfitting,a layer of Transformer encoder is added downstream of the BERT model;Fi⁃nally,a four channel dual layer CNN model is used to improve the performance of analyzing news comment sentiment by combining convolu⁃tional kernels of different sizes.The experimental results show that the accuracy of this method on two news comment datasets reaches 93.0%and 96.4%,respectively;The comparative experiments with different models further demonstrate the effectiveness of the proposed method.

关 键 词：情感分析 BERT模型 Transformer Encoder 多尺度CNN 新闻评论 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]