基于双向长短期记忆模型的网民负面情感分类研究

Negative Emotions of Online Users' Analysis Based on Bidirectional Long Short-Term Memory

网民负面情感在网络舆情情感分析中具有重要意义,但已有研究缺乏自动化识别海量短文本中网民负面情感的多分类方法。本文利用词嵌入技术学习词语的特征表示,通过增加文本的情感特征生成具有情感意义的词向量,并训练双向长短期记忆模型得到网民负面情感识别模型,在判断网民情感极性的基础上,识别网民的愤怒、悲伤和恐惧三种负面情感,并结合案例数据与SVM、LSTM和CNN等模型进行对比分析。实验表明,具有情感语义的词向量比词向量更适合情感分析任务;利用双向长短期记忆模型可以得到较好的情感识别效果;判断网民情感极性基础上识别网民负面情感的分类方式优于直接判断网民的负面情感的方式。

The negative emotions of netizens are of great significance in the analysis of network public opinions; however, existing research lacks a multiclassification method to identify automatically the negative emotions of online users in massive short text. This study uses word embedding to study the features of the word sequence, learning the sentiment-encoded word vector by increasing the emotional features of the context. The bidirectional long short-term memory（LSTM） model is trained to obtain the online users＇ negative-emotion recognition model, and the online users＇ anger, sadness, and fear are identified on the basis of judging the online users＇ sentiment polarities. Then, the case data are compared with experimental results of models such as support vector machine（SVM）, LSTM, and convolutional neural network（CNN）. The experimental results show that sentiment-encoded word embedding is more suitable than word embedding for sentiment analysis. The bidirectional long short-term memory model has good sentiment analysis performance. Classifying the negative emotions of the netizen based on the identification of sentiment polarities is better than directly distinguishing negative emotions.