Social Spam Discovery Using Bayesian Network Classifiers Based on Feature Extractions

People always communicate with each other through social networking services (SNSs). However they often receive various kinds of unwelcomed messages that can be requests from uncomfortable friends or may be advertisements. In this paper, we defined these messages as "social spams", and suggested new classification method to detect them. By characterizing the problem of discovering social spams which frequently occurs in current popular SNSs, we extracted and exploited novel features that had not shown in the existing E-mail or web spamming prevention techniques. Our proposal for collecting various features such as behavior, celebrity, trust, common interest, etc. could incrementally been updated for SNS users. We modified the existing well-known classification techniques such as Bayesian network classifiers (BNCs) to customize for SNS features. To make decision efficiently, we computed Katz or trust scores with only part of updated network topologies.