International Research Journal of Education and Technology

This project develops a hybrid deep learning framework for emotion detection in social media posts, combining CNNS and LSTMS to enhance accuracy by capturing local and sequential features in text data. Pre- trained embeddings like Word2Vec and GloVe are used for better semantic understanding, while attention mechanisms and Transformers improve model performance. The framework integrates facial expression recognition through CNNS, creating a multi-modal approach for robust emotion detection. Implemented in Python using TensorFlow and Keras, the model is evaluated on diverse, multi-lingual datasets collected via APIs and web scraping tools from various social media platforms. Standard metrics such as accuracy, precision, recall, and Fl-score are used for evaluation. The project aims to address challenges like slang and informal language, ensuring the model is adaptable and effective across diverse social media environments.

Key Words: Hybrid deep learning, Emotion detection, Social media posts, Attention mechanisms, Transformers, Multi- modal approach, LSTM, Word2Vec, GloVe, CNN.