HSSD: Hate speech spreader detection using N-Grams and voting classifier

Profane or abusive speech with the intention of humiliating and targeting individuals, a specific community or groups of people is called Hate Speech (HS). Identifying and blocking HS contents is only a temporary solution. Instead, developing systems that are able to detect and profile the content polluters who share HS will be a better option. In this paper, we, team MUCIC, present the proposed Voting Classifier (VC) submitted to Hate Speech Spreader Detection shared task organized by PAN 2021. The task includes profiling HS spreaders for two languages, namely, English and Spanish from the text collected from Twitter. This task can be modeled as a binary text classification problem to classify an author (Twitter user) based on his/her tweets as 'Hate speech spreader' or 'Not'. The proposed models utilizes a combination of traditional char and word n-grams with syntactic ngrams as features extracted from the training set. These features are fed to a VC that employs three Machine Learning (ML) classifiers namely, Support Vector Machine (SVM), Logistic Regression (LR), and Random Forest (RF) with hard and soft voting. The proposed models with accuracies of 73% and 83% for English and Spanish languages respectively, obtained second rank in the shared task.