From Tweets to Feelings: A Novel GPT-3 and BiLSTM Fusion for Social Media Emotion Detection
Mengling Wang, Qun Hou, Ao Peng · 2024
The surge in social media usage underscores the critical need for precise emotion extraction from voluminous texts. This study introduces a novel model that leverages the advanced capabilities of GPT-3 and BiLSTM for enhanced accuracy and efficiency in detecting emotions within social media texts. Our methodology employs two comparative approaches: assessing each model's performance under consistent training parameters and optimizing each model's performance via targeted parameter tuning. Results indicate that the fusion model outperforms the conventional single model across both experimental conditions, particularly in terms of accuracy and F1-scores. Experimental findings demonstrate that our fusion model attains superior performance in accuracy, recall, and F1-scores on the Sentiment140 dataset compared to the conventional single model. This research highlights the applicability of fusion models in social media sentiment analysis and offers critical insights for future deep learning model design and optimization.