Sentiment analysis on social media

• Main Author: Wang, Hai
• Other Authors: Mao Kezhi
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Sentiment analysis; Deep learning; GLOVE embedding; Emoji
• Online Access: https://hdl.handle.net/10356/175974
• Institution: Nanyang Technological University

This research investigates improving sentiment analysis by integrating pre-trained GloVe embeddings with LSTM and Transformer architectures. Driven by the need for finer sentiment interpretation in digital communication, we evaluated models to determine GloVe embeddings’ impact on sentiment classification accuracy. Our approach included processing a varied dataset, recognizing emojis as crucial sentiment indicators, and using advanced neural networks to more effectively capture sentiment expression nuances. Results showed significant accuracy improvements in sentiment analysis with GloVe embeddings, especially with the Transformer model in interpreting context-rich text. Importantly, the study highlighted emojis’ critical role in enhancing both LSTM and Transformer models’ sentiment classification capabilities. These outcomes highlight the promise of merging deep learning methods with semantic embeddings to refine sentiment analysis, providing key insights for applications from social media analytics to monitoring customer sentiment. This research’s main contribution is the thorough comparative analysis of LSTM and Transformer models with GloVe embeddings, establishing an extensive framework for utilizing deep learning in sentiment analysis. This foundation encourages further exploration into multi-modal sentiment analysis and the creation of advanced models for comprehensively understanding human emotions in text.