Authors:
Shengyuan Wen
and
Weiqing Sun
Affiliation:
College of Engineering, University of Toledo, Toledo, Ohio, U.S.A.
Keyword(s):
Social Spam, Online Learning, Incremental Learning, Data Scraping, Semi-supervised Learning, Label Spreading.
Abstract:
Social network users receive a large amount of social data every day. These data may contain malicious unwanted social spams, even though each social network has its social spam filtering mechanism. Moreover, spammers may send spam to multiple social networks concurrently, and the spam on the same topic from different social networks has similarities. Therefore, it is crucial to building a universal spam detection system across different social networks that can effectively fend off spam continuously. In this paper, we designed and implemented a tool Spam-Fender to facilitate spam detection across social networks. In order to utilize the raw social data obtained from multiple social networks, we utilized a semi-supervised learning method to convert unlabelled data into usable data for training the model. Moreover, we developed an incremental learning method to enable the model to learn new data continuously. Performance evaluations demonstrate that our proposed system can effectively
detect social spam with satisfactory accuracy levels. In addition, we conducted a case study on the COVID-19 dataset to evaluate our system.
(More)