Publication Date

Spring 2024

Degree Type

Master's Project

Degree Name

Master of Science in Computer Science (MSCS)


Computer Science

First Advisor

Robert Chun

Second Advisor

Navrati Saxena

Third Advisor

Kamal Penmetcha


Emotion Detection, Deep Learning, Mental Health, NLP, BERT, BiLSTM, TBED, Ensemble, Multi Label Classification


Emotion detection is gaining exponential necessity in today’s technological age. This research seeks to delve into ways conversational AI could be enhanced by integrating emotional intelligence using an ensemble learning approach. Traditional machine learning along with advanced neural network architectures are implemented to improve the understanding and intricacies of emotion detection from textual data. The dataset we use is GoEmotions dataset, annotated with 27 emotional labels, to conduct a detailed analysis of emotion recognition. Various machine learning models, such as HistGradientBoosting, LightGBM, CatBoost, and MLP, will be evaluated side by side with advanced models of Bidirectional Long Short-Term Memory (BiLSTM) and Bidirectional Encoder Representations from Transformers (BERT). Ensemble learning techniques that aggregate individual models' predictive capabilities can significantly improve accuracy, underlining the efficacy of integrated methodologies. Although traditional models provide a very robust foundation, we aim to analyse if the BERT and BiLSTM models can overcome the shortcoming of shallow ML models and comprehensively capture the complexity of human emotions. The success of the ensemble model can help us foresee the method's applicability in advancing the sophistication of AI systems that demand sensitive emotion recognition. Data from different domains is tested on the final model to challenge its efficacy. This research aims to articulate a path toward developing more empathetic and context-aware AI interactions. This report further articulates future directions: the incorporation of multimodal data, dealing with data imbalance, optimization for real-time processing, and improvement of cross-lingual and cultural adaptability will all serve further to enhance the robustness and applicability of emotion detection systems.

Available for download on Saturday, May 24, 2025