Hybrid Deep Learning-Based Approach for Enhanced Multi-Label Emotion Classification

Nowadays, social networking is controlling our lives. Every passing second, millions of tweets are shared by individuals through online social media platforms like Twitter as a means to express their feelings and thoughts. Exploring and examining the emotions conveyed within social media content, along with extracting valuable insights from extensive data, categorizing it into various groups, and foreseeing end-user behavior or emotions, holds potential advantages across multiple domains such as public health, social welfare, commerce, etc. This paper offered a deep learning-based solution for addressing multiple emotion classification challenges on Twitter. The novel proposed presents an advanced deep learning system with a hybrid approach that harnesses the strengths of combining Bidirectional Long-Short Term Memory and Recurrent Neural Network with an additional attention mechanism. This model has been trained using the "SemEval 2018 Task1 (Twitter Data)" to evaluate the performance of the suggested model. The experimental outcomes demonstrated that, compared to individual SVM, RNN, or LSTM models, as well as hybrid models, it produces classification results that are more accurate and have greater F1 scores and recall. Our model surpasses the existing cutting-edge systems, achieving a superior accuracy score of 0.90. The proposed model provides an alternate resolution to the issue of long-term dependency within current models while also addressing the challenge of data loss with growing training data sizes. The presented model achieved improved accuracy, which was amplified further with larger training data sizes as well as an increased number of training epochs.