DETECTING AND MITIGATING ADVERSARIAL INFLUENCE OPERATIONS IN NETWORKS
Abstract
Malicious social manipulation is a fast growing threat to US national interests. Our adversaries use targeted social media attacks to spread propaganda and conspiracy theories that disrupt the normal function of our society and promote distrust in our institutions. Increasingly, malicious actors are also weaponizing our social platforms by exploiting vulnerabilities in the algorithms they use to highlight important content. This and other coordinated attacks allow malicious actors to scale up influence operations to skew the perceptions of entire populations. The evolving nature of social manipulation requires us to adapt by developing new defensive capabilities. The goal of the proposed work is to develop such capabilities by identifying populations susceptible to targeted attacks and reducing their vulnerability. First, proposed work will elucidate how network structure, jointly with algorithmic curation of content by the platform, shapes the information individuals see in online networks. We will identify vulnerable populations that receive biased content and quantify their susceptibility. To reduce this bias, we will create mitigation strategies and validate them empirically. Proposed work will also develop methods to identify distributed influence campaigns where multiple actors coordinate to promote some information. To address the challenges of inherent noise and partial observability, we will extend the unsupervised changepoint detection method to automatically identify the start of coordinated operations. Additionally, we will efficiently label malicious accounts so as more accurately detect influence campaigns. Resulting methods will robustly detect evolving adversarial manipulation in online networks and track it over time. The methods will identify populations susceptible to social manipulation, and devise appropriate mitigation strategies that help protect against social manipulation.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 12, 2021
- Source ID
- FA95502010224
Entities
People
- Kristina Lerman
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- University of Southern California