Sentiment Analysis Model Based on Structure Attention Mechanism

Abstract

Since the long short-term memory (LSTM) network is a sequential structure, it is difficult to effectively represent the structural level information of the context. Sentiment analysis based on the original LSTM causes a problem of structural level information loss, and its capacity to capture the context information is finite. To address this problem, we proposed a novel structure-attention-based LSTM as a hierarchical structure model. It may capture relevant information in the context as much as possible. We propose HM (h_t matrix) to storage the structural information of the context. Furthermore, we introduce the attention mechanism to realize vector selection. Compared with the original LSTM and normal attention-based sentiment classification methods, our model obtains a higher classification precision. It is proved that the structure-attention-based method proposed in this study has an advantage in capturing the potential semantic structure.