潘梅

摘要：为帮助政府等相关部门及时掌握大众对特定公共事件的主要情感倾向，针对基于词向量的深度学习方法实现网民情绪识别，存在高度依赖分词准确性、一词多义等问题，提出基于BERT-BiLSTM的网民情绪识别方法。首先，基于BERT预训练模型获取预处理后的待识别文本词向量;然后，利用BiLSTM提取上下文相关特征进行学习;最后，通过分类器获得文本的情感极性，包括积极和消极两类。通过对疫情期间网民情绪识别数据集实验表明，基于BERT-BiLSTM的网民情绪识别模型P值为88.98%，R值为92.72%，F1值为90.81%，相比于LSTM和BiLSTM模型性能更优。本识别方法可为网民情绪识别研究提供借鉴，识别结果可为政府决策分析和舆情引导提供参考。

关键词：网民;情绪识别;BERT;BiLSTM

中图分类号：TP391        文献标识码：A

文章编号：1009-3044（2021）18-0074-03

开放科学（资源服务）标识码（OSID）：

Emotion Recognition of Internet Users based on BERT-BiLSTM

PAN Mei

（Chengdu Normal University， Chengdu 611130， China）

Abstract： In order to help the government and other relevant departments grasp the main emotional tendencies of the public on specific public events in time， aiming at the problems of high dependence on word segmentation accuracy and polysemy in deep learning method based on word vector to realize internet users emotion recognition， this paper proposes ainternet users emotion recognition method based on BERT-BiLSTM. Firstly， the text word vector with preprocessed and to be recognized is obtained based on the BERT; secondly， the context sensitive features are extracted by the BiLSTM for learning; finally， the emotional polarity of the text， including positive and negative， is obtained through the classifier. The experiments of internet users emotion recognition data set during the epidemic period shows that the Precision is 88.98%， Recall is 92.72%， and F1 is 90.81%based on BERT-BiLSTM， which is better than LSTM and BiLSTM. The recognition method can provide reference for the research of internet users emotion recognition， and the recognition results can provide reference for government decision-making analysis and public opinion guidance.

Key words： internet users; emotion recognition; BERT; BiLSTM

1 引言

随着互联网技术和移动通信技术的高速发展，普通大众均可便捷地在网络上对公共事件发表评论和分享观点，通过网络表达对各种社会事件的情感态度。该方式可以迅速传播和延伸至网络各区域，引发公众关注和热议，形成网络社会舆论，网络舆论通常会产生巨大的舆论动向和影响。公共事件爆发后，政府须尽快掌握人们对该话题的关心程度和发展趋势，有效提高其公信力和应急管理能力。如新型冠状病毒（COVID-19）感染的肺炎疫情，迅速引发国内和国际舆论的持续关注，众多网民参与疫情相关话题讨论。政府部门需要掌握公众在该事件传播过程中的情感状况和社会舆论情况，科学高效地做好防控宣传和舆情引导工作。……