Stop the spark detecting violence instigating Bangla text using ViolenceBanglaBERT

Abstract Due to the low cost and widespread availability of the internet and digital devices, the utilization of social networks is rapidly increasing. Therefore, in recent years, a group of individuals from different religions and cultural backgrounds has used social network platforms to promote misinformation and hatred, which has instigated communal violence and devastating consequences. Regarding this concerning rise of communal violence instigated through social media texts, this study presents a comprehensive benchmarking and optimization framework of multiple deep learning and transformer-based models for violence-instigating Bangla text classification. A wide range of embedding methods, deep learning architectures, transformer-based models, and existing approaches are systematically evaluated under a unified experimental framework. To address class imbalance, data augmentation via back-translation is applied to improve generalization. The experimental results illustrate that the proposed ViolenceBanglaBERT model, a fine-tuned version of a transformer-based classifier, achieved a Macro F1-score of 0.75 on the test set, outperforming other classifiers and existing approaches. The source code used in this study can be accessed at https://github.com/prosenjit-ch/SS-VIBTI .