Detecting Disseminators of Hate Speech by Optimizing Textual and User-Based Features

Hate speech through online social media is on the rise, which has attracted researchers across the world to develop models to control its circulation to create better and smarter society. Although several models have been reported to recognize hate speech in the most widely spoken language, the issue has not yet been deemed to be under control. Very few academics have built models to identify hate speech spreaders so that the problem can be managed at the source by correctly filtering their posts. In this article, we employ long short-term memory (LSTM) to extract features from user-posted text-tweet and Tweepy to extract behavior-based features. The LSTM-based textual feature extracted from text-tweet and user behavior based features extracted from the Twitter user’s profile are concatenated and further optimized with camel algorithm (CA) to yield more pertinent features. The optimized feature vector is used to determine whether or not a user is a hate spreader using machine learning classifiers. The developed model is validated with the dataset provided by the PAN 2021 and PAN-2020 (https://pan.webis.de/data.html) hate speech spreader profiling and FakeNews spreader profiling challenge, respectively. Our proposed model outperforms the state-of-the-art models, with an accuracy of 96% for PAN 2021 and 94% for PAN 2020 for English and Spanish datasets, respectively.