Enhancing Text Sentiment Classification Through RoBERTa-Based NLP Models

Abstract

With the vast amount of user-generated content on social media and online platforms, sentiment analysis has emerged as an important research field in the domain of Natural Language Processing (NLP). Although moderate success has been obtained by traditional machine learning or deep learning approaches in text sentiment classification, they are not very good in capturing the semantics in the context, the sarcasm and the informal language patterns used in social media text. To overcome these drawbacks, this research introduces a RoBERTa based approach for sentiment classification, which leverages transformer-based contextual embeddings to better understand and classify sentiment. The proposed methodology includes data pre-processing, Byte-Pair Encoding tokenization, transfer learning and fine-tuning of a pretrained model RoBERTa on a balanced sentiment dataset of textual and sentiment samples. Standard performance metrics such as accuracy, precision, recall and F1-score are used to evaluate the model. The experimental results presented show that the proposed framework using RoBERTa outperforms the conventional NLP methods in terms of sentiment classification performance and interpretation ability. RoBERTa's bidirectional attention mechanism, optimized pretraining strategy, allows for efficient processing of semantic relationships, social media sentiment and noisy text. The study also examines the pros and cons of sentiment analysis systems based on transformers and their potential for future development. Overall, the research proves that RoBERTa is a very powerful and reliable model for the current sentiment classification problems and can be a major step towards developing intelligent opinion mining and automated text analysis systems.