Social Media Botnet Detection with Linkage Analysis and Machine Learning
summary
Claims a behavior‑based linkage graph plus graph clustering and label‑propagation finds coordinated botnets at scale, not just topic signals. Pipeline builds pairwise bot linkage from IP/user‑agent/profile/activity, runs graph clustering (GraphX‑style) and a semi‑supervised propagation step, ingesting logs with Kafka and Spark. Pinterest reports over half a billion accounts flagged historically (≈30% of all accounts), and one detected campaign yields a cluster of 54k nodes and 900 pages that lets analysts block or shadowban and patch rules.
AI-written summary · about
chapters (10)
- 0:00Intro
- 1:41Outline of Today's Talk
- 2:46Rise of Social Media
- 3:42Social Media Spam
- 7:08Spam Botnet
- 9:57Case Linkage Analysis
- 14:52Unsupervised Learning
- 17:16Semi-supervised Learning
- 22:20Machine Learning Pipeline
- 24:21Applications
related talks
23mBot or human? Detecting malicious bots with machine...
44mApplied Machine Learning for Data Exfil and Other Fun Topics
36mDetecting Influence Campaigns Across Social Media Platforms
25mTesting Your Organization's Social Media Awareness
21mMachine Learning and Cyber Security - Detecting malicious URLs in the haystack
48mBehind the Scenes: The Industry of Social Media Manipulation Driven by Malware
36mA Graph-based Machine Learning early warning system to detect Ransomware
48mWho’s in control: Human or Machine?