Emotion Recognition in Text using BiLSTM and Deep Bidirectional Graph Convolution Networks

In the rapidly developing field of natural language processing (NLP), emotion recognition from text has become more crucial for better human-computer interactions, particularly in chatbots and sentiment analysis. In this work, Deep Bidirectional Graph Convolutional Networks (DBGCN) and Bidirectional "Long Short-Term" Memory (BiLSTM) are combined to create a hybrid deep learning method for emotion recognition. The approach uses BiLSTM to capture the text's sequential dependencies and contextual subtleties, while DBGCN improves feature learning by take into consideration the links between various text characteristics. The recommended model handles text data by first transforming it into numbers via "Term Frequency-Inverse Document Frequency" (TF-IDF) vectorization, and then classifying emotions using the learned features. Experimental results show that the suggested technique significantly improves accuracy and robustness, beating traditional models such as ANN and SVM. The model was trained and evaluated using a custom dataset that contained a mix of Twitter and Kaggle datasets, as well as synthetically created data, yielding a comprehensive representation of 20 distinct emotions. The algorithm had an incredible 93% accuracy in emotion classification and emoji production. This study demonstrates the effectiveness of BiLSTM and DBGCN for emotion identification, underlines the relevance of emoji output in increasing user interactions, and shows the model's appropriateness for real-time applications.