Sentiment Analysis Based on Hybrid Bi-attention Mechanism in Mobile Application

Sentiment analysis is one of the fundamental tasks in nature language processing field, as well as in mobile application. The transformation of message text information into Emoji display can improve interactive experience, but there is a lack of specific introduction to the transformation process. On the other hand, Deep Learning has achieved great process in text sentiment classification, e.g. LSTM and bi-LSTM, however, the existing LSTM models ignore the backward information and bi-LSTM models ignore the interaction information when calculating the forward and backward features independently. To address these issues, we propose a novel hybrid bi-attention (HBA) neural network to capture the forward, backward and bi-direction information simultaneously. Then, we also design a combine strategy to train these three part information. The experimental results show that our proposed hybrid bi-attention model achieves better performance in sentiment analysis, and the constructed emotional display system can automatically turns the message text into an emoji picture display.