Twitter Data Based on Airline Feedback Using BERT (Bi-Directional Encoder Representations from Transformers)

The multifaceted nature of words, logical requirements, and the presence of noise in social media text make it difficult to sense the sequence of textbased information, particularly in tweets pertaining to carriers. Traditional sentiment analysis techniques that rely on TF-IDF or word frequency techniques usually miss subtle contextual meanings and connotations in the text. The proposed system improves the accuracy of opinion arrangement by integrating effective preprocessing techniques with deep learning-based models. To ensure a robust understanding of emotion in tweets, proposed a hybrid deep learning model that combines BERT for feature extraction and MLP for sequential pattern recognition. The final configuration is carried out using softmax activation and fully connected layers. This paper significantly improves the accuracy of sentiment classification by employing contextual word embeddings and sequential learning, which makes it suitable for real-time airline sentiment analysis and customer feedback monitoring. The comparative results on implementation of Naive Bayes, SVM, LSTM, MLP+Word2Vec and BERT+MLP are analysed. The performance measures in terms of accuracy, precision, recall and f1-score shows comparatively better in BERT+MLP algorithm.