A method for detecting abnormal users with fake stars

In GitHub, users star interesting repositories, and the number of stars is viewed as the significant measure of repository popularity. Some repositories obtain fake stars by unjustified means, which ruin efforts that communities have made stars a valuable indicator, and bring negative impacts in GitHub. Therefore, it is important to stop abusing GitHub stars and detect abnormal users who provide fake stars. In this paper, we first define features from the user dimension and repository dimension. Then we perform differential analysis and find that most of the features show a significant difference between abnormal users and normal users. Next, we propose a method AUDetec for Abnormal User Detection. The method AUDetec uses the decision tree to detect the abnormal users based on two features, including the sum of repositories starred by the user and the median value of the number of days since creation for repositories starred by the user. We evaluate the effectiveness of AUDetec on the data set which contains 120 abnormal users and 240 normal users. The experiment results show that AUDetec has a high performance by achieving an accuracy of 99.86% on average.