EmoNet: Unveiling Affective States Through Convolutional Neural Networks in Textual Emotion Classification

In the burgeoning field of natural language processing, emotion classification from textual data has emerged as a critical task with applications ranging from sentiment analysis to mental health assessment. This paper explores the utilization of Convolutional Neural Networks (CNNs), traditionally dominant in image processing, for classifying emotions in text. Our proposed CNN model leverages the inherent hierarchical structure of language to identify and learn emotion-specific features, with an emphasis on capturing contextual n-grams through convolutional filters. The approach is substantiated by a comprehensive dataset, subjected to rigorous preprocessing and vectorization via TF-IDF to convert text into a numerical format suitable for deep learning. The model's architecture is meticulously crafted, incorporating convolutional layers followed by global max pooling and dense layers, culminating in a softmax activation function tailored for multi-class classification. Our findings demonstrate the model's robustness, achieving a notable accuracy of 96.08% on the test set. This high level of precision is further corroborated by the Receiver Operating Characteristic (ROC) analysis, revealing exceptional area under the curve (AUC) values across various emotion categories. The results suggest that CNNs hold significant promise for emotion recognition tasks in textual data, providing an effective framework for future explorations in the domain.