Comparison of RNN and LSTM Classifiers for Sentiment Analysis of Airline Tweets

Abstract

This study examines the application of Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) models for sentiment analysis of airline-related tweets, focusing on customer feedback directed at U.S. airlines on the X platform (formerly Twitter). The objective was to utilize these deep learning models to identify sentiment trends within text data and compare their performance in terms of computation time. The analysis was conducted on a 14,640-imbalanced dataset of classified tweets from February 2015 as positive, negative, or neutral. Both models were trained under identical conditions using Term Frequency-Inverse Document Frequency (TF-IDF) and Word2Vec for feature extraction. LSTM achieved 74% accuracy with AUC scores of 0.84, 0.90, and 0.89. RNN achieved 72% accuracy with AUC scores of 0.78, 0.87, and 0.85. In terms of time efficiency, RNN outperformed LSTM, requiring 57.16 seconds for training and 0.52 seconds for testing, compared to LSTM’s 82.40 and 0.82 seconds. Time performance was also evaluated per sentiment class, and RNN consistently outperformed LSTM. These results highlight the trade-off between accuracy and computational cost. Limitations include dataset imbalance and LSTM’s slower computation due to its internal gate mechanisms. Future work could prioritize integrating hybrid models and may use data imbalance techniques to improve sentiment classification.