Year: 2025
Type: dissertation
Abstract: This thesis investigates the effectiveness of different embedding techniques for emotion classification in natural language text. Using the GoEmotions dataset, we evaluate multiple embeddings including contextual and dimensionality-reduced variants across various classification models such as BERT, Bi-GRU, and logistic regression. The study focuses on a multi-label emotion detection task, where each emotion is treated as an independent binary classification problem. Performance is assessed using standard evaluation metrics, and emotion-specific threshold selection is employed for improved prediction. Our findings highlight the impact of embedding choice on classification accuracy and provide insights for building more robust emotion-aware NLP systems.
Authors: Kolli, Maheedhar, Kolli, Maheedhar
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7080238382
