Text Emotion Detection Based on Bi- LSTM Network
Zihe Wang · Academic Journal of Computing & Information Science · 2020
Emotion detection and opinion mining in social network is a hot research topic nowadays. Most existing studies and research, however, have been using typical binary (positive/negative) or ternary (positive / negative / neutral) classification. Classifying short sequences of text into multi subclasses is relatively rarely reported. Except Bouazizi and Ohtsuki (2017), the accuracy rate of detection in their two studies was only 56.9% and 60.2%. In our study, we proposed a Bi-directional Long Short-Term Memory with Language Model (BiLSTM-LM) to classify text sequences into seven distinct emotional classes. Results showed that the accuracy rate of detection based on our model can reach as high as 64.09% on multi-class classification, which is 4 percentage points higher than the most advanced model in the world to date.