AN APPROACH TO IDENTIFY VULGARISM BASED ON MACHINE LEARNING

The article develops an approach to the identification of vulgarism in web content based on machine learning. The increasing number of harmful contents in web-pages makes protection from them even more vital. Encountering vulgarisms (indecent words, jargon, slams, etc) on the internet among users, especially children, and teenagers, shows a negative effect on their psychology. To identify vulgar words, word conjunctions, and expressions, in both social media (Twitter, Facebook, etc.) and online media, it is important to develop new auto-text identification methods, which is vital to solve that matter. The presented paper proposes an approach for the detection of vulgarisms using the N-grams+TF-IDF features. Numerical vectors are generated by applying the n-gram+TF-IDF-based feature extraction method to the predefined vulgar words. Generated numerical vector is passed to the input of the Naive Bayes algorithm. As a result of experiments conducted on different features, the classification based on unigram+TF-IDF features performs better results. The proposed approach, which contains the identification of vulgarism, is important for developing conversation culture and communication skills of children and teenagers. This approach is very important to protect kids from harmful content online and can be used in child safety centers and education systems (pp.89-98).