Enhanced Sentiment Analysis of Airline Twitter Review Using Hybrid Machine Learning and Deep Learning Models

Understanding public opinions and trends requires effective sentiment analysis of social media data. This research explores advanced approaches for sentiment classification using the Twitter US Airline Sentiment Dataset, which contains 14,640 tweets. We assess the effectiveness of various models, including Long Short-Term Memory (LSTM) networks, Naive Bayes, and Support Vector Machines (SVM), along with different text vectorization techniques: Bag of Words (BoW), TF-IDF, and Word2Vec. The results demonstrate that LSTM networks achieve an impressive accuracy of 94.5%, surpassing SVM (90%) and Naive Bayes (85.6%). Among vectorization methods, both BoW and TF-IDF yield the highest AUC scores at 0.92, while Word2Vec-based techniques provide improved F1-Scores, with W2V-AVG at 0.66 and W2V-TFIDF at 0.65. These findings highlight the superior performance of LSTM and Word2Vec for accurate and balanced sentiment classification.