Comparative Analysis of LSTM and BiLSTM for Sexual Harassment Classification on Indonesian Social Media

Sexual harassment on social media is a growing concern, especially on platforms that allow unrestricted user interactions. This study aims to compare the performance of two deep learning models, Long Short-Term Memory (LSTM) and Bidirectional Long Short-Term Memory (BiLSTM), in classifying sexually harassing comments on Indonesian-language social media. The comparison is intended to identify which model performs better in accurately detecting such harmful content. To address the class imbalance, the dataset was processed using the Synthetic Minority Over-sampling Technique (SMOTE). Both models were then evaluated based on accuracy, precision, recall, and F1-score metrics. The results indicate that, while both models achieved high classification accuracy, the LSTM model slightly outperformed BiLSTM with an accuracy of 94%, precision of 96%, recall of 94%, and F1-score of 94%. In contrast, the BiLSTM model attained an accuracy of 93%, precision of 95%, recall of 93%, and F1-score of 94%. These findings suggest that LSTM has a marginal advantage over BiLSTM in this classification task, making it a potentially more effective choice for detecting sexually harassing comments. This comparison provides valuable insights into model selection for automated detection of harmful content on Indonesian-language social media, contributing to efforts to create a safer digital environment.