HateClassify: A Service Framework for Hate Speech Identification on Social Media
Muhammad Usman Shahid Khan, Assad Abbas, Attiqa Rehman, Raheel Nawaz
Abstract
It is indeed a challenge for the existing machine learning approaches to segregate the hateful content from the one that is merely offensive. One prevalent reason for low accuracy of hate detection with the current methodologies is that these techniques treat hate classification as a multiclass problem. In this article, we present the hate identification on the social media as a multilabel problem. To this end, we propose a CNN-based service framework called “HateClassify” for labeling the social media contents as the hate speech, offensive, or nonoffensive. Results demonstrate that the multiclass classification accuracy for the CNN-based approaches particularly sequential CNN (SCNN) is competitive and even higher than certain state-of-the-art classifiers. Moreover, in the multilabel classification problem, sufficiently high performance is exhibited by the SCNN among other CNN-based techniques. The results have shown that using multilabel classification instead of multiclass classification, hate speech detection is increased up to 20%.