Fine-Grained Emotion Classification

Sentiment analysis usually means positive, negative, or neutral. GoEmotions is harder: 27 fine-grained emotions where the boundaries blur, comments carry more than one feeling, and the classes are heavily imbalanced.

This semester project built a text-classification pipeline on the dataset, working through tokenisation, model choice, and the evaluation needed to tell whether the model has learned anything beyond predicting the most common labels.

It was a hands-on study in where modern NLP models struggle, which is rarely the easy, well-separated cases.