Comparative Evaluation of LSTM, BiLSTM, and CNN Algorithm for Sentiment Analysis of Tinder Application

This study investigates the performance of three deep learning architectures Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), and Convolutional Neural Network (CNN) for sentiment classification of user reviews from the Tinder application in Indonesia. A dataset of 5,000 recent reviews collected from the Google Play Store was utilized. The research adopts the Knowledge Discovery in Databases (KDD) framework, encompassing data selection, preprocessing, transformation, data mining, and evaluation stages. Experimental results show that, in baseline evaluation, the CNN model achieved the highest accuracy of 98%. After hyperparameter optimization, LSTM improved its performance to 98.2%, while BiLSTM outperformed all models with an accuracy of 98.7%. The application of class weighting significantly enhanced the models’ sensitivity to minority classes, resulting in more balanced and reliable predictions. These findings demonstrate that deep learning models, when combined with hyperparameter tuning and imbalance handling techniques, can achieve highly accurate and robust sentiment classification. Furthermore, this study highlights the effectiveness of BiLSTM in capturing contextual dependencies in user-generated text, making it particularly suitable for sentiment analysis in online dating application reviews. The results provide valuable insights for developers and stakeholders to better understand user satisfaction and improve service quality.