Cyberbullying identification in twitter using support vector machine and information gain based feature selection

Abstract

Cyberbullying is one of the actions that violate the ITE Law where the crime is committed on social media applications such as Twitter. This action is difficult to detect if no one is reporting the tweet. Cyberbullying tweet identification aims to classify tweets that contain bullying. Classification is done using Support Vector Machine method where this method aims to find the dividing hyperplane between negative and positive class. This study is a text classification where more data is used, the more features are produced, therefore this research also uses Information Gain as feature selection to select features that are not relevant to the classification. The process of the system starts from text preprocessing with tokenizing, filtering, stemming and term weighting. Then perform the information gain feature selection by calculating the entropy value of each term. After that perform the classification process based on the terms that have been selected, and the output of the system is identification whether the tweet is bullying or not. The result of using SVM method is accuracy 75%, precision 70.27%, recall 86.66% and f-measure 77.61% on experiment maximum iteration = 20, λ = 0.5, γ = 0.001, ε = 0.000001, and C = 1. The best threshold of information gain is 90%, with accuracy 76.66%, precision 72.22%, recall 86.66% and f-measure 78.78%.