Detection of Multiple Emotions in Texts Using Long Short-Term Memory Recurrent Neural Networks

Abstract

Recognizing emotions from text applies to every part of our lives, like enhancing human-computer interaction, mental health monitoring, recognizing public sentiment about any national, international, or political event. Given the importance of emotion analysis, especially the classification of multi-labeled emotions, this paper proposes a deep learningbased system to address the issue of classifying multi-labeled emotions in texts. Toward this aim, by combining several datasets, a dataset first created which all samples are multilabeled, and then, using the long short-term memory recurrent neural network (LSTM), a new network is designed to detect multiple emotions from the texts. The GloVe and FastText have been used to find semantic, syntactic, and related words. Moreover, the attention property is utilized to improve the accuracy of the network. The comparative results indicated that the proposed model performs better compared to the existing methods in terms of accuracy.