Improving Prediction Accuracy of Emotions from X (Twitter) Post Opinions with Enhanced Hierarchical Attention Model

Emotion detection from X (Twitter) posts has garnered considerable interest due to its potential applications in sentiment analysis and social media monitoring. The information morality of tweets makes it difficult to recognize emotion on the platform, including slang, abbreviations, and emoticons. Traditional machine learning methods often struggle with this complexity, prompting the exploration of deep learning techniques. Artificial neural networks are the foundational model used, followed by neural networks with recurrent architectures (RNNs) and their variants. Convolution neural networks are utilized as well to extract spatial features, even though long-term memory and gated recurrent neural networks are better at simulating long-term connections in sequential data. This method elevates the ability of the model to extract meaningful features from tweets while capturing temporal dynamics. The current study demonstrates the effectiveness of deep learning models, especially the proposed hierarchical attention model, for X (Twitter) emotion detection. Achieving a 97.25% accuracy indicates the potential for real-world applications for social media monitoring, sentiment analysis, and consumer feedback analysis. The proposed approach has a mean accuracy of 99.95%, 99.96%, 99.94%, and 99.95%, according to cross-validation employing 3, 5, 7, and 10 folds. Comparison with existing approaches also corroborates its superiority and robustness.