Комплексний підхід до виявлення деструктивного українськомовного контенту в соціальних медіа

Abstract

This paper presents developing and validating a complex approach to classify content into constructive and destructive. The fine-tuned FacebookAI/xlm-roberta-base transformer model achieved the best-balanced accuracy classification score of 93.45%. Our developments can be used in conjunction with each other to provide moderate resource consumption and improve interpretability.