Advancing Text Emotion Recognition: BERT and BiLSTM Integration

Social media platforms have become rich sources of emotional expression, offering a plethora of textual and visual data for emotion detection and analysis. To identify emotions on social media, users' shared text, photos, and videos are analyzed using NLP. Text-based emotion recognition on social media frequently encounters difficulties because of things like informal slang, irony, and the widespread usage of emoticons and emojis. A hybrid model that combines bidirectional LSTM layers with BERT embeddings for text-based emotion identification is proposed. The inherent difficulties in deciphering and interpreting emotional nuances in textual data, especially from noisy sources like social media are addressed by this combination. With the help of Bi-LSTM's capacity to record temporal relationships in sequential data and BERT's strength in feature extraction and contextual understanding, the proposed framework provides improved textual input emotion identification accuracy. Five-fold cross-validation offers an extensive evaluation of the model's performance on several data subsets. The model is demonstrated by consistent performance gains and high accuracy metrics seen during cross-validation. The hybrid model for text-based emotion recognition produces better results in terms of accuracy, adaptability, and robustness.