Social Spammer Detection Based on Semi-Supervised Learning

Abstract

With the rapid development of Internet and mobile communication technologies, the social media has been spreading into all aspects of people's life, work and study. People can not only communicate with each other and share news through social media platforms, but also check news and information and learn about popular topics. However, due to the social platform's characteristics, such as rapid information dissemination and interaction mechanism among users, it has also become the main attacking surface for adversaries. Spammers mainly use social media platforms to spread phishing, fraud, publish pornography, malicious content and links to make profits, which seriously disrupts the normal operation of social media platforms and causes adverse effects on society. In this paper, we combine the semi-supervised model with maximum contrastive pessimistic likelihood (MCPL) estimation and the ensemble learning CatBoost algorithm, proposing a semi-supervised ensemble learning classification algorithm model (SSML-CatBoost) for spammer detection in social media, and comparative experimental results show that our model outperforms other models for different amounts of labeled data and different numbers of iterations.