Effects of Feature Extraction and Classification Methods on Cyberbully Detection

Cyberbullying is defined as an aggressive, intentional action against adefenseless person by using the internet, or other electronic contents. researchershave found that many of the bullying cases have tragically ended in suicides; henceautomatic detection of cyberbullying has become important. in this study we showthe effects of feature extraction, feature selection, and classification methods thatare used, on the performance of automatic detection of cyberbullying. to performthe experiments formspring.me dataset is used and the effects of preprocessingmethods; several classifiers like c4.5, naïve bayes, knn, and svm; and informationgain and chi square feature selection methods are investigated. experimentalresults indicate that the best classification results are obtained when alphabetictokenization, no stemming, and no stopwords removal are applied. using featureselection also improves cyberbully detection performance. when classifiers arecompared, c4.5 performs the best for the used dataset.