Comparative Analysis of CNN-LSTM and LSTM Models for Cyberbullying Detection with Increasing Dataset Sizes
Tri Pratiwi Handayani, Mohamad Ilyas Abas · Jurnal Ilmu Komputer (JUIK) · 2024
This study compares the performance of two deep learning models, CNN-LSTM and LSTM, for identifying cyberbullying in social media text. Three distinct dataset sizes are used for our evaluation and comparison: 1,000, 5,000, and 10,000 samples. The results indicate that the CNN-LSTM model outperforms the LSTM-only model (Ablation model) for the largest dataset size, exhibiting substantial enhancements in accuracy, precision, recall, and F1-Score as the dataset size increases. The Ablation model exhibits competitive performance and slightly superior results on the mid-sized dataset. However, it inevitably falls behind the CNN-LSTM model when trained on 10,000 samples. These findings imply that increasing the complexity of the CNN layer in the CNN-LSTM model improves its ability to collect significant features in bigger datasets, making it more successful for cyberbullying detection.