Automatic detection of influential actors in disinformation networks

Area of Science:

• Computer Science
• Social Science
• Network Science

Background:

• Disinformation campaigns on digital platforms pose significant challenges to cybersecurity.
• Hostile influence operations (IOs) leverage social media for widespread dissemination.
• Identifying and countering these operations requires advanced detection methods.

Purpose of the Study:

• To develop an end-to-end framework for automating the detection of disinformation narratives, networks, and key actors.
• To quantify the impact of individual actors in spreading influence operation narratives.
• To provide a robust system for identifying malicious social media activity.

Main Methods:

• Integration of natural language processing (NLP), machine learning (ML), and graph analytics.
• Application of network causal inference to assess actor impact.
• Testing on real-world datasets from the 2017 French presidential elections and extensive Twitter data (2007-2020).

Main Results:

• The framework achieved 96% precision and 79% recall in detecting influence operation (IO) accounts.
• Identified salient network communities and high-impact accounts missed by traditional metrics.
• System performance validated against independent sources including US Congressional reports and investigative journalism.

Conclusions:

• The developed framework offers a powerful, automated solution for detecting and analyzing large-scale disinformation campaigns.
• It effectively identifies influential actors and network structures within hostile influence operations.
• This approach significantly enhances the ability to counter online disinformation threats.