The rapid growth of Online social media (OSM) platforms has led to an explosion of user-generated content rich in text emotional expression, making emotion detection from text an essential task in Natural Language Processing (NLP). However, accurately detecting nuanced emotions from short, informal, and often ambiguous text remains a big challenge, especially for traditional sentiment analysis models. This paper aims to advance emotion recognition by enhancing the capability of deep learning approaches in capturing complex emotional signals in social media text. We propose and evaluate two deep learning models, Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (BiLSTM), trained on the AdamCodd dataset. GloVe 6B 300d pre-trained word embeddings were used to enhance semantic representation, and model performance was tested using accuracy, precision, recall, and F1-score. The BiLSTM model outperformed the CNN, achieving a total accuracy of 92% compared to 91% for the CNN, and both models significantly outperformed traditional machine learning baselines. These findings demonstrate that context-aware deep learning models, specifically sequence-based architectures such as BiLSTM, are highly suitable for emotion classification in short, informal text and establish the groundwork for developing emotionally intelligent systems in areas like mental health, marketing, and user interaction.