Robust Hybrid Model for Social Media Sentiment Analysis

Social media platforms are home to massive amounts of content created by users, which can offer insightful information on attitudes, beliefs, and feelings. In order to extract valuable information from the abundance of data, sentiment analysis is essential. This work presents a novel hybrid sentiment analysis model using the Bidirectional Encoder Representations from Transformers (BERT) and Robustly Optimized BERT Approach (RoBERTa) models. The strategy tries to improve sentiment classification performance by utilizing the advantages of both models. The complexities of social media text are often difficult for traditional sentiment analysis models, like rule-based systems and machine learning algorithms like logistic regression and support vector machines, to capture because of issues with contextual understanding, linguistic complexity, inherent data variability, and noise. This hybrid sentiment analysis, on the other hand, excels at understanding natural language semantics and gathering contextual information. Labelled data and an innovative training approach are combined to train the model. Reporting accuracy metrics is part of the performance evaluation process when using a test dataset. Additionally, a thorough classification report and confusion matrix are produced. A test dataset performance evaluation yields an accuracy of 82%.