摘要
[研究目的]舆情评论情感分析是帮助相关部门及时掌握网民诉求、合理疏导舆情的重要抓手。为解决传统文本分析模型无法准确判别掺杂反讽语义文本的情感极性问题,设计了一种协同双向编码表征模型。[研究方法]将两个普通双向编码表征模型协同组合,分别进行反讽语义/非反讽语义、正面情感/负面情感的语义理解能力训练。然后将获取的反讽识别向量与情感识别向量通过一个额外的全连接层进行合并,构建协同双向编码表征模型。在反讽识别向量的指导下,此模型会根据评论文本的不同性质,在输出层进行不同的对应处理。[研究结论]以“望江女子溺水案”为例进行实验,结果表明:与普通双向编码表征、Text-CNN和Text-LSTM模型相比,协同双向编码表征模型的P、R、A、F1等指标均有明显提高。且在此基础上进行的LDA主题挖掘,可实现舆情评论情感极性的主题可视化,为相关部门进行舆情管控提供更加精准的决策支持。
[Research purpose]Sentiment analysis of public opinion comments is an important tool to help relevant departments respond to netizens'demands in time and guide public opinion reasonably.In order to solve the problem that traditional text analysis models cannot accurately identify the emotional polarity of irony doped text,the Collaborative BERT was designed.[Research method]The two BERTs were combined to train the semantic comprehension ability of irony/non-irony and positive/negative emotions respectively.Then,the obtained irony recognition vector and emotion recognition vector were combined through an additional full connection layer to construct the Collaborative BERT.Under the guidance of the irony recognition vector,the model will conduct different corresponding processing in the output layer according to the different nature of the comment text.[Research conclusion]Taking the Drowning Case of Wang-jiang Woman as an example,the results showed that compared with the BERT,TEXT-CNN and TEXT-LSTM models,the indexes of P,R,A and F1 of the Collaborative BERT were significantly improved.On this basis,LDA theme mining can realize the topic visualization of the polarity of public opinion comments,and provide more accurate decision support for relevant departments to control and guide public opinion.
作者
潘宏鹏
汪东
刘忠轶
李轲
Pan Hongpeng;Wang Dong;Liu Zhongyi;Li Ke(School of Police Administration,People's Public Seourity University of China,Beijing 100076;PLA Rocket Force NCO College,Weifang 262500)
出处
《情报杂志》
CSSCI
北大核心
2022年第5期99-105,111,共8页
Journal of Intelligence
关键词
舆情
反讽识别
协同双向编码表征
情感分析
public opinion
ironic identification
collaborative BERT
sentiment analysis
