An approach to detect abusive content incorporating Word2Vec and Multilayer Perceptron

With the rapid growth of social media text, millions of negative comments are flowing on social webs and social networking sites. Abusive content is harmful to people and societies that can provoke various criminal offenses like hate crimes. Hate speech is also a form of abusive content. An automatic and improved detection system for hate speech can help to reduce this problem. Implicit abusive content requires contextual semantic and syntactical analysis. We propose a novel abusive text detection model with the word2vec model and compositional vector model to analyze text more semantically and syntactically. The proposed model considers the English language dataset for abusive text. The abusive content detection model exhibits achievable performance compare to various deep learning and machine learning classifiers. Among all models, Multilayer Perceptron classifier achieves 86% accuracy compared to other models.