Analyzing File-to-File Relation Network in Malware Detection

摘要: Due to its major threats Internet security, malware detection is of great interest both the anti-malware industry and researchers. Currently, features beyond file content are starting be leveraged for e.g., file-to-file relations, which provide invaluable insight about properties samples. However, we still have much understand relationships benign files. In this paper, based on relation network, design several new robust graph-based reveal relationship characteristics. Based designed two findings, first apply Malicious Score Inference Algorithm MSIA select representative samples from large unknown collection labeling, then use Belief Propagation BP algorithm detect malware. To best our knowledge, investigation characteristics network in using social analysis. A comprehensive experimental study a sample relations obtained clients software Comodo Security Solutions Incorporation performed compare various approaches. Promising results demonstrate that accuracy efficiency proposed methods outperform other alternate data mining techniques.