基于BERT的日常文本情感分析  

Daily text sentiment analysis based on BERT

作　　者：张泽源 张光妲 李佳雨 王海珍[1] ZHANG Ze-yuan;ZHANG Guang-da;LI Jia-yu;WANG Hai-zhen(College of Computer and Control Engineering,Qiqihar University,Heilongjiang Qiqihar 161006,China)

机构地区：[1]齐齐哈尔大学计算机与控制工程学院,黑龙江齐齐哈尔161006

出　　处：《齐齐哈尔大学学报（自然科学版）》2024年第6期37-41,共5页Journal of Qiqihar University(Natural Science Edition)

基　　金：大学生创新创业训练计划项目(S202310232142,x202310232014,x202310232016,S202310232116);教育部供需对接就业育人项目(20230104647,20230103670);黑龙江省教育厅基本业务专项(145209313);黑龙江省规划办项目(GJB1421345);齐齐哈尔大学教育科学研究项目(GJZRYB202030,GJQTZX2021001,GJQTZX2021027,GJQTYB202152,GJQTYB202206);齐齐哈尔大学教学课程研究项目(QDJG_2022KCSZ16);教育部协同育人项目(202102019011,201901022009,201901148011,202002155036)。

摘　　要：针对常见文本中情绪倾向性不明显的问题,提出一种融合BERT模型与网络爬虫技术的应对方法。利用网络爬虫技术搜集社交网络的数据资料为情感分析提供实例和数据支撑,采用交错验证法将数据集分割为训练子集和预测子集。利用数据集进行BERT学习和文本分类,即运用深度学习和自然语言处理对文字资料实施归类。研究结果表明,采用BERT模型实行情感分析的技术在精确度与广泛适用性上展现出良好的性能,正确率可达97.86%。For the problem that the emotional tendency is not obvious in common texts,this paper proposes a coping method combining BERT model and web crawler technology.The use of web crawler technology to collect data from social networks provides examples and data support for emotion analysis.Interleaved validation was used to split the datasets into training and prediction subsets.Using data sets for BERT learning and text classification,that is,deep learning and natural language learning are used to classify text data.The study showed that the use of BERT model to implement emotion analysis showed good performance in accuracy and wide applicability,with an accuracy rate of 97.86%.

关 键 词：BERT 情感分析 自注意力机制 预训练模型 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]