Improving Fine-Grained Emotion Detection in Text with BERT and GoEmotions: An Experimental Study

Emotion detection in text is critical for applications ranging from mental health monitoring to human-computer interaction. While traditional machine learning models struggle with nuanced emotional expressions, transformer-based architectures such as BERT (Bidirectional Encoder Representations from Transformers) offer promise due to their contextual understanding. This study explores the effectiveness of fine-tuning BERT on the GoEmotions dataset, a large-scale corpus of 58000 Reddit comments labeled with 27 emotion categories. We propose a streamlined pipeline that leverages transfer learning to adapt BERT for multi-label emotion classification. Our experiments demonstrate that the fine-tuned BERT model achieves an accuracy of 85% and an F1-score of 0.83, outperforming baseline models such as Support Vector Machines (SVMs) (72% accuracy) and Long Short-Term Memory (78% accuracy). The model excels in distinguishing subtle emotions (e.g., “gratitude” versus “joy”) but faces challenges with semantically overlapping categories such as “sadness” and “disappointment.” These results highlight the potential of contextual embeddings for emotion recognition while underscoring the need for more robust datasets. This work contributes to the development of emotionally intelligent AI systems, with applications in personalized chatbots, mental health diagnostics, and sentiment-aware content moderation.