Sentiment Analysis on Social Media Data using Roberta-GRU Hybrid Model

Sentiment analysis, a key technique natural language processing (NLP) determines the tone of emotion in text and express as positive or negative or any other personal feeling. The explosion of social media content has made sentiment analysis indispensable across various fields, including e-commerce, healthcare, politics and digital marketing. However, social media data provide unique analytical challenges due to its informal language, inherent noise, misspellings and the prevalence of code-mixed text. Traditional sequential models often fall short in addressing these complexities, demanding more robust and adaptable solutions. This paper introduces an enhanced RoBERTa-GRU hybrid model specifically designed for improved sentiment classification on noisy social media datasets. Our approach incorporates text normalization, tokenization and fine-tuning of RoBERTa embeddings to better capture the nuances of informal language. The incorporation of Gated Recurrent Units (GRU) enhances the model's capacity to capture sequential dependencies present in text, thereby improving the accuracy of sentiment classification. We assess the proposed model using real-world datasets characterized by noise, specifically consisting of comments from Twitter and reviews from YouTube. Preliminary findings indicate that our RoBERTa-GRU hybrid model significantly surpasses conventional approaches, achieving an accuracy rate of 90%. These outcomes highlight the model's proficiency in addressing the complexities associated with noisy, usergenerated content. Our results emphasize the promise of this hybrid methodology in advancing sentiment analysis within social media contexts, facilitating more precise and contextually pertinent insights.