A Multilingual Offensive Language Detection Method Using Transfer Learning from a Fine-Tuned Transformer Mode

The swift growth of global internet communication between various languages and cultures has increased the need for efficient multilingual offensive language identification systems. Modern analytical models mostly deal with single-language contexts which leaves fundamental gaps in spotting obstacles connected to the multilingual nature of present digital systems. The research creates an approach that delivers dependable solutions to identify hostile language across multiple linguistic environments. Using modern natural language processing techniques, the main objective focuses on correct identification of objectionable information across different languages which includes both low-resource and underrepresented languages. The method allows the model to solve problems from insufficient annotated datasets in many languages while adapting its pre-trained contextual representations specifically for abusive language identification tasks. Research data is obtained from multiple sources through the use of Kaggle datasets including Jigsaw Multilingual Toxic Comment Classification. The data must undergo text encoding and tokenization operations together with low-resource language management. The offensive language identification model receives its final shape by applying the transfer learning properties of multilingual transformer models. Evaluation tests prove the suggested model achieves higher performance than standard neural architectures and machine learning models. The model surpasses baseline systems by a significant margin on multilingual offensive language tests with 98.95% accuracy together with 98% F1-score. The research develops a complete new multilingual hostile language detection method by applying transfer learning to fine-tune previously trained multilingual transformer models. These results confirm the robustness of the proposed system by revealing its high efficiency in securing different language online exchanges.