Selective shallow models strength integration for emotion detection using GloVe and LSTM

Abstract

Text analysis has gained immense popularity due to the widespread use of the internet and unrestricted access to people’s opinions provided by social media. Analyzing public emotions in real-time can enable us to predict problematic situations like civil unrest that may arise in the future allowing us to take measures to prevent or handle them. This paper proposes a novel technique for emotion detection that can be used in real-time due to its comparatively much smaller run time and smaller memory size. Present well-performing models for emotion detection are incapable of being used in real-time due to the incorporation of large deep learning models that make them considerably slower. This work proposes a technique to use multiple shallow models to surpass the performance of a single large model by selectively combining their strengths and disregarding their weaknesses. These shallow models work independently which allows them to be run in parallel to ensure a smaller execution time. This combined proposal achieved 86.16% accuracy in 00.98 milliseconds per input. Therefore, the experiments show that the proposed model outperforms state-of-the-art models. Moreover, the computational cost shows that the proposal may used for real time applications.