casprlab's repositories
AbuSniffV1
Adversaries leverage social network friend relationships to collect sensitive data from users and target them with abuse that includes fake news, cyberbullying, malware, and propaganda. Case in point, 71 out of 80 user study participants had at least 1 Facebook friend with whom they never interact, either in Facebook or in real life, or whom they believe is likely to abuse their posted photos or status updates, or post offensive, false or malicious content. We introduce AbuSniff, a system that identifies Facebook friends perceived as strangers or abusive, and protects the user by unfriending, unfollowing, or restricting the access to information for such friends. We develop a questionnaire to detect perceived strangers and friend abuse. We introduce mutual Facebook activity features and show that they can train supervised learning algorithms to predict questionnaire responses.
DSG
A densest subgraph based solution to partition co-review graphs into worker controlled components.
LBP
Loopy Belief Propagation (LBP) approach that formulates the problem of detecting fraudulent user accounts as a network classification task.
MCDense
An algorithm that partitions a co-review graph into components, each controlled by a different worker. MCDense exploits the observation that the accounts controlled by a worker are likely to be used to review many products in common, while two accounts controlled by different workers are unlikely to have reviewed many products in common.