Attention based hierarchical LSTM network for context-aware microblog sentiment classification

Analyzing sentiment polarities of microblog has become a hot research topic for both academic and industrial communities. Most of the existing algorithms regard each microblog as an independent training instance. However, the sentiments embedded in short tweets are usually ambiguous and context-aware. Even a non-sentiment word might convey a clear emotional tendency in certain microblog conversations. In this paper, we regard the microblog conversation as sequence, and develop a Context Attention based Long Short-Term Memory (CA-LSTM) network to incorporate preceding tweets for context-aware sentiment classification. The CA-LSTM network has a hierarchial structure for modeling microblog sequence and allocates the words and tweets with different weights using attention mechanism. Our proposed method can not only alleviate the sparsity problem in feature space, but also capture long distance sentiment context dependency in microblog conversations. Experimental evaluations on a public available dataset show that the proposed CA-LSTM network with context information can outperform other strong baselines by a large margin.