A Comparative Study of LSTM Models on Sentiment Analysis

In this paper, we conduct a comparative analysis of LSTM (Long Short-Term Memory) architectures, including GPU LSTM and Convolutional LSTM, for sentiment analysis tasks on an e-commerce dataset. The evaluation is based on precision, recall, and F1-score metrics for classifying positive and negative sentiments in product reviews. Both GPU LSTM and Convolutional LSTM models demonstrate strong performance in classifying positive sentiments. Moreover, these models achieve high precision, recall, and $F 1$-score. However, they struggle with negative sentiment classification, exhibiting zero precision, recall, and F1-score for this class. Overall, both models achieve an accuracy of 0.97, indicating their effectiveness in sentiment analysis tasks. Furthermore, our study underscores the importance of further research to enhance the ability of the LSTM model to classify negative sentiments in e-commerce datasets accurately.