Context-Aware Chinese Microblog Sentiment Classification with Bidirectional LSTM

摘要: Recently, with the fast development of microblog, analyzing sentiment orientations tweets has become a hot research topic for both academic and industrial communities. Most existing methods treat each microblog as an independent training instance. However, sentiments embedded in are usually ambiguous context-aware. Even non-sentiment word might convey clear emotional tendency conversations. In this paper, we regard conversation sequence, leverage bidirectional Long Short-Term Memory (BLSTM) models to incorporate preceding context-aware classification. Our proposed method could not only alleviate sparsity problem feature space, but also capture long distance dependency Extensive experiments on benchmark dataset show that LSTM context information outperform other strong baseline algorithms.