A hybrid based model on LSTM-CNN to multi-class emotion analysis on social networking dataset

Abstract

Sentiment analysis from texts has been a major research eld in NLP. However, most of the studies are on binary (positive and negative) classi cation of the texts. While researching, we found that the accuracy of multi-class text classi cation according to emotions is very low when compared to binary classi cations, as understanding and quantifying emotions is a very di cult task. We studied the two commonly used deep learning models used for text classi cations: Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN). We found that the greatest accuracy was achieved when the CNN model is used combined with a LSTM. In our paper, we proposed an LSTM-CNN hybrid model to classify texts according to ve emotion classes and achieve an accuracy of 65%. We further studied Support Vector Machine (SVM) and Naive-Bayes classi ers. The experimental results show that the LSTM-CNN model had an improved accuracy.