Comparison of LSTM and TCN for Sentiment Analysis in Supporting Churn Prediction

The growth of e-commerce in Indonesia has driven an increase in transactions and changes in digital shopping behavior. However, challenges such as customer churn remain a significant issue affecting business sustainability. Sentiment analysis has become an essential method for understanding customer experiences and identifying potential churn. This study compares two deep learning methods, Long Short-Term Memory (LSTM) and Temporal Convolutional Network (TCN), in sentiment analysis of customer reviews on e-commerce platforms. Review data from Baliyoni Group (Balimall and Tokodaring) undergoes filtering, cleaning, tokenization, normalization, and labeling before being used to train the LSTM and TCN models. Evaluation results show that the TCN model has a slight advantage with an accuracy of 86.81%, while LSTM achieves 85.73%. Although the accuracy difference is not significant, TCN is more stable in identifying negative and neutral sentiments. Expert validation indicates a high level of validity for both models, making these methods relevant for e-commerce applications and churn reduction strategies.