Detecting Offensive Posts on Social Media

Xiao Chen, Xinfeng Ye, Manoranjan Mohanty, Sathiamoorthy Manoharan

Abstract: While social media platforms bring people closer together and facilitate the rapid spread of information, they also face the challenge of offensive and harmful content, such as bullying, discrimination, and hate speech. Most recent research has focused on detecting text-based offensive language in social media. However, audio-based detection methods are still underdeveloped. This paper addresses this gap by proposing a novel deep learning-based model, CLS-CNN, for identifying objectionable audio. We collect a video dataset for training machine learning models for detecting offensive language. We compare the performance of CLS-CNN with traditional machine learning methods using the collected dataset. Results showed that the CLS-CNN model achieved an accuracy of 88% on our collected dataset and outperformed other models. The model was also evaluated using three publicly available text-based datasets. It shows that CLS-CNN achieves at least a 2% accuracy improvement compared to the other published schemes using the same dataset.