A team of researchers from Australia and Bangladesh has developed an advanced machine learning model that can detect toxic comments with 87% accuracy, significantly improving automated content moderation. This new model, created through collaboration between East West University and the University of South Australia, outperforms previous methods by reducing false positives and improving classification reliability.
The system was trained on a dataset containing English and Bangla comments gathered from various online platforms. After evaluating three different models, the optimised Support Vector Machine (SVM) model emerged as the best performer, surpassing both a baseline SVM model (69.9% accuracy) and a Stochastic Gradient Descent model (83.4%).
As online interactions grow exponentially, manual moderation becomes increasingly unsustainable, and automated tools are necessary for efficient content management. Many existing models struggle with false positives, but the new SVM-based model addresses these issues by offering higher accuracy and more effective content filtering, while minimizing the risk of removing legitimate posts.
The research team is continuing to refine the model, focusing on deep learning integration and expanding the dataset to include multiple languages and regional dialects to further enhance its applicability across diverse online communities. The goal is to develop a system that can be deployed in real-world online platforms, offering a more scalable and reliable solution to the growing problem of online toxicity.
In the future, the team plans to improve the model’s processing time, adapt it to evolving online behaviors, and ensure its integration into content moderation workflows on digital platforms. By leveraging advancements in AI and machine learning, they aim to create more sophisticated tools to foster safer, more respectful online interactions.
