New AI model detects toxic comments with an accuracy of 87.6%

Researchers from Australia and Bangladesh have developed an innovative machine learning model that can detect toxic comments on social media with an accuracy of 87.6%.

Amazon Prime Video DOWNLOAD

This advance was presented at the 2024 conference on Innovation and Intelligence in Computing, where the growing concerns about cyberbullying and its devastating effects on the mental health of online users were discussed.

The combination of efforts between East West University in Bangladesh and the University of South Australia has resulted in an algorithm that outperforms existing models, which often generate a high number of false positives.

An AI to solve a serious problem

The principal investigator, data science expert Afia Ahsan, highlighted that the increase in cyberbullying has led to serious problems such as self-harm and even suicide in extreme cases.

“Eliminating toxic comments from social media platforms is vital to curb the rising abuse and ensure respectful interactions in the social space,” Ahsan stated.

The optimized model was tested on a dataset containing comments in English and Bangla, collected from prominent platforms such as Facebook, YouTube, and Instagram.

The team found that their support vector machine (SVM) model was the most reliable and effective among three tested models, making it the preferred option for real-world implementation.

Researchers plan to enhance their capabilities by integrating deep learning techniques and expanding the dataset to include more languages and regional dialects.

Amazon Prime Video DOWNLOAD

On the other hand, partnerships with online platforms are being explored to implement this technology, which could transform the way we interact in the digital realm and significantly contribute to the improvement of mental health on social media.