Team_Syrax at BLP-2023 Task 1: Data Augmentation and Ensemble Based Approach for Violence Inciting Text Detection in Bangla
Omar Riyad, Trina Chakraborty, Abhishek Dey · 2023
This paper describes our participation in Task 1 (VITD) of BLP Workshop 1 at EMNLP 2023, focused on the detection and categorization of threats linked to violence, which could potentially encourage more violent actions.Our approach involves fine-tuning of pre-trained transformer models and employing techniques like self-training with external data, data augmentation through back-translation, and ensemble learning (bagging and majority voting).Notably, self-training improves performance when applied to data from external source but not when applied to the test-set.Our analysis highlights the effectiveness of ensemble methods and data augmentation techniques in Bangla Text Classification.Our system initially scored 0.70450 and ranked 19th among the participants but post-competition experiments boosted our score to 0.72740.