Spam User Detection Interactive Machine Learning System with Spam User Seeds on Social Media

Abstract

There has been globally an increase in the use of SNS such as Twitter, Facebook, and Instagram. Also, as much as various information are offered in a web page, there are a lot of information that discomforts many users. Hence, as many cases that someone hacks SNS users frequently occur due to the increased use of social media, there are a lot of hacked IDs used for unwholesome advertisements such as a sale of IDs and illegal gambling in Twitter, Facebook, and Instagram, and advertisement controlling systems are very unstable. This paper calculated the access idle time between the date when general users composed tweets and the date when the hacked users composed tweets in order to decide whether users’ SNS accounts like Twitter were hacked or not and identified patterns of users’ general tweets and the hacked tweets for advertising. Furthermore, we observed the hacked users’ followings and decided whether they were hacked or not. Using Trust Lank Algorithm, we tried to examine spam users by investigating top-K rank IDs from the hacked IDs’ followings to Trust Rank. Moreover, we visualized general users’ tweets by using a tag cloud and verified the effectiveness of algorithms presented by a comparison with other algorithms suggested in this study. We also realized an interactive machine learning interface that accepts system users’ feedback to improve the performance of a system and verified that the relevant system’s performance of examining spam users was very excellent compared to a baseline through experiments.