Emotion Classification of Short Social Media Texts Using Classical Machine Learning and TF-IDF Features

This paper presents a machine learning approach to classifying tweets into four distinct emotions: anger, fear, joy, and sadness. Utilizing the WASSA-2017 shared task dataset, the study evaluates classical natural language processing (NLP) approaches that combine bag-of-words and TF-IDF features with various classifiers, including Naive Bayes, Support Vector Machines (SVM), Logistic Regression, Random Forests, and Gradient Boosting.The research provides a transparent, efficient, and reproducible baseline for multi-class emotion detection in short social media texts. It demonstrates the effectiveness of ensemble methods, particularly Gradient Boosting, while discussing the limitations of using default hyperparameters, unigram features, and a restricted four-emotion label set. The paper concludes by outlining future directions, such as incorporating word embeddings, fine-tuned transformer models (e.g., BERT), emoji/hashtag features, and expanding the label set to include additional emotions like surprise, disgust, and love.