Zarindokht Helforoush ^* Thu Thu Hlaing

• Florida Institute of Technology, Melbourne, United States

Understanding public sentiment on social media platforms like Twitter and Facebook has become increasingly crucial in various domains, including marketing, politics, and customer service. Social media serves as a rich source of real-time data, providing insights into public opinion towards products, services, events, and social issues. This study focuses on enhancing sentiment analysis on different social datasets using advanced machine learning techniques. In this research, we present a new approach for sentiment analysis, leveraging a hybrid model combining Particle Swarm Optimization with BiLSTM neural networks. The proposed hybrid PSO-BiLSTM model aims to optimize key parameters of the BiLSTM architecture, including the number of hidden layers, neurons, and training epochs, to enhance the predictive accuracy of sentiment analysis tasks.We conducted experiments on three benchmark datasets and compared the performance of the proposed model with standard BiLSTM and other existing approaches. The results demonstrate that the hybrid model consistently outperforms baseline models, achieving higher accuracy rates in predicting text sentiment. Through the adaptive tuning facilitated by PSO algorithm, the proposed model demonstrates its effectiveness in optimizing BiLSTM parameters, leading to improved performance across diverse datasets. This research contributes to advancing sentiment analysis techniques, providing a robust framework for automatic opinion discernment with practical implications for various domains.