Enhancing Twitter Sentiment Analysis using Attention-based BiLSTM and BERT Embedding

Sentiment analysis is a rapidly expanding field with a broad spectrum of uses. Researchers and industrialists alike have found analyzing sentiments on social media platforms like Twitter, to be of particular interest due to the influx of opinionated data. In this paper, we propose an Attention-based BiLSTM sentiment model for Twitter data that is integrated with BERT embedding. The BERT pre-trained language model represents each word as a vector, while the Bi-Directional Long Short Term Memory (BiLSTM) extracts word information from both directions. To enhance prediction accuracy, the attention mechanism determines how much each word contributes to the final score. We conducted experiments using the Sentiment140 dataset and evaluated the results based on ac-curacy, recall, precision, and Fl-Score. The empirical results reveal that the pro-posed model outperforms the baseline model. Our model effectively analyzes and interpret the vast amount of opinionated data on Twitter providing valuable in-sights for researchers and businesses alike.