基于双重注意力模型的微博情感倾向性分析

Microblog sentiment orientation analysis based on double attention model

在现有的微博情感倾向性分析任务中,微博标签往往被视为噪声信息,在数据预处理阶段就被剔除.但微博标签蕴含着微博内容的关键信息,所以标签的剔除对于微博的情感倾向性分析是不利的.针对该问题,充分考虑微博的文本特点,提出一种基于双重注意力的情感分析模型.采用Bi-LSTM(Bi-directional Long Short-Term Memory)分别构建微博文本和微博标签的语义表示,采用双重注意力机制同时对微博的正文层和微博的标签层进行语义编码,提取出文本中的关键信息.最后,基于所构建的语义表示训练情感分类模型.实验结果表明,该模型在微博情感倾向性分析上取得了较好的效果.

In the existing microblog emotion orientation classification task,the microblog tag is often regarded as noise information and eliminated in the data preprocessing stage. But the microblog tag contains key information and the elimination of tags is not good for microblog emotional orientation classification. Aiming at this problem,this paper fully considers the characteristics of microblog text and proposes a sentiment analysis model based on double attention model. The Bi-LSTM(Bi-directional Long Short-Term Memory)is used to construct the semantic representation of the microblog text and the microblog tag. The double attention mechanism is used to semantically encode the microblog text and the microblog tag and to extract the key information in the text. Finally,the sentiment classification model is trained based on the constructed semantic representation. Experimental results show that the model achieved good results in the microblog emotional sentiment analysis.