融合TCN与BiLSTM+Attention模型的疫情期间文本情感分析

Text sentiment analysis during the epidemic based on TCN and BiLSTM+Attention fusion model

鉴于目前主流的文本情感分析方法存在难以解决长期依赖和对上下文信息使用不足的缺陷,本文首次提出将时序卷积网络(TCN)和BiLSTM+Attention模型融合的文本情感分析模型。该模型利用TCN的因果卷积和扩张卷积结构获取更高层次的文本序列特征,并通过双向长短期记忆网络(BiLSTM)进一步学习上下文相关信息的情感特征;最后,引入自注意力机制(Self-Attention)帮助模型优化特征向量,提高情感分类的准确度。在新型冠状病毒疫情期间的微博文本数据集上进行对比实验,结果表明该模型的性能相较于其它模型有明显的提升。

At present,there are some difficulties in solving the problems of long-term dependence and the insufficient use of contextual information in the mainstream text sentiment analysis methods.In this paper,a text sentiment analysis model combining Temporal Convolutional Network(TCN)with BiLSTM+Attention model is proposed for the first time.This model uses TCN's causal convolution and extended convolution structures to obtain higher-level text sequence features,and further learns the emotional features of contextual information through the Bidirectional Long Short-Term Memory network(BiLSTM).Finally,the Self-Attention mechanism(Self-Attention)is added to optimize the feature vectors of this model and improve the accuracy of sentiment classification.A comparative verification experiment is conducted on the Weibo text data set during the novel coronavirus epidemic.The results show that the performance of this model is more significantly improved than that of other models.