Automated Detection of Image-Based Abusive Bengali Memes Using Ensemble and Deep Learning Models

In the realm of internet communication, memes have emerged as a powerful means of expression, swiftly disseminating emotions and ideas. The ability to classify memes, particularly in languages with regional nuances such as Bengali, poses unique challenges. The primary question of this study addresses how to efficiently classify Bengali memes, given their diverse and context-specific nature. The motivation for this research stems from the growing significance of memes as a form of internet communication and the need for effective methods to categorize and understand them. This work presents here a computational model of classifying Bengali memes using convolutional neural networks. Convolutional neural networks have the potential to learn, adapt, and rearrange themselves. Along-side employing traditional machine learning models-Support Vector Machine (SVM), Logistic Regression, and Random Forest for meme classification. The experiment illustrated a CNN model with Bi-LSTM layers is implemented, achieving a test accuracy of 69.49%. Additionally, Among the machine learning models, the Random Forest achieves the highest accuracy of 64%. Our study shows that this comparison between CNN-based and traditional models provides valuable insights for future research in meme analysis and content moderation.