C Vishwakarma, N Singh and S Srivastava
Sentiment analysis, of social media data presents significant challenges due to its informal and noisy nature of the data, which often includes slang, misspellings, emojis and abbreviations. Traditional sentiment analysis methods struggle to accurately interpret such information. This paper proposes a RoBERTa-GRU hybrid model designed to improve sentiment analysis in noisy social media data. The study begins by identifying the limitations of existing methods in handling noisy, informal text. The proposed solution integrates text normalization techniques to preprocess the data and effectively manage informal language features. RoBERTa embeddings are fine-tuned on noisy datasets, enhancing the model's ability to understand and adapt to the complexities of social media text. Additionally, the inclusion of GRU models enables the capture of sequential dependencies and temporal context, crucial for interpreting sentiment evolution in dynamic social media interactions. The methodology explores useful datasets from platforms like Twitter and YouTube, where the model's effectiveness in handling noisy data is tested. The paper concludes by assessing the model’s strengths and weaknesses in real-world applications showing how the hybrid approach improves sentiment classification accuracy in challenging noisy environments.