Twitter Buzzer Detection for Indonesian Presidential Election

Abstract

The campaign that was done in social media has high correlation to the supporters who disseminating the information deliberately, which called as buzzer. However, data that were generated by buzzer accounts can be considered as noise and need to be removed. In this research we performed task for detecting the buzzer accounts in Twitter by observing the impact of features we used which we selected based on their Mutual Information scores. We examined the performance of four machine learning algorithms which are Ada Boost (AB), Gradient Boosting (GB), Extreme Gradient Boosting (XGB), and Histogram-based Gradient Boosting (HGB). The algorithms were evaluated using 10 folds cross validation and the results show that the best accuracy and precision achieved by AB which are 62.3% and 61.3% respectively with 25 features while the recall attained by XGB (67.9%) which the score same with its recall result with 20 features.