Coordinated influence operations on social media leave measurable traces in the timing and structure of user interactions, according to a new computational study published in EPJ Data Science. Researchers Simone Mungari, Emilio Ferrara and Luca Luceri analyzed temporal interaction networks from five real-world influence campaigns attributed to state-sponsored actors from China, Cuba, Iran, Russia and the United Arab Emirates, and built two complementary tools: one to identify the accounts driving these campaigns, and another to predict which organic users will interact with them in the future.
The study addresses two interconnected questions: how can influence operation drivers be distinguished from legitimate users, and can future interactions between those drivers and organic users be predicted? The authors frame their work as a response to a critical threat to the integrity of public discourse, where early detection and accurate prediction of user engagement remain open challenges.
Before building their models, the team first established why detection matters. In four of the five campaigns studied, they found empirical evidence that organic users’ adoption of content pushed by influence operations, such as hashtags and URLs, was statistically associated with prior exposure to the activity of influence operation actors. This is an association observed in the data, not a demonstrated causal effect, but it links exposure to later uptake of campaign content among ordinary users.
The first tool, called IOBoost, is a lightweight supervised classifier that identifies influence operation driver accounts. It relies on three families of signals: temporal features capturing when accounts act, activity-based features describing how they behave, and network motif features reflecting the structural patterns of their interactions. Across the campaigns, the classifier achieved an average F1-score of 97.5 percent, with strong cross-campaign generalization and the ability to detect drivers early in a campaign’s life.
The second tool, IOLink, is a graph neural network model that works on user interaction networks to predict future interactions between influence operation drivers and organic users. It achieved an average ROC-AUC score of 86 percent, meaning it ranks future driver-organic interactions substantially better than chance across the tested campaigns.
The authors report that coordinated influence campaigns leave distinctive behavioral traces both in user adoption patterns and in temporal interaction dynamics. According to the study, these signatures enable accurate and generalizable detection while remaining computationally efficient enough for real-world deployment, a consideration that matters because platforms must process interaction data at very large scale.
To support verification and reuse, the researchers made their source code publicly available on GitHub, and the article is published open access. The paper appeared on 7 September 2026 and is marked as an early-shared, peer-reviewed accepted version that carries a permanent DOI and remains citable, with a final version of record to follow.
The scope of the findings is defined by the data: five campaigns attributed to state-sponsored actors in five countries. The reported performance figures are averages across these campaigns, and the association between exposure and content adoption was observed in four of the five rather than all of them. The study presents a detection and prediction framework and its measured performance on these datasets; it does not report deployment results on live platforms or establish causal mechanisms by which exposure changes user behavior.
Within those limits, the work offers a concrete technical recipe for platforms and researchers: characterize accounts by how their activity unfolds over time and how their interactions are structured, use those features to flag likely campaign drivers, and model the interaction network itself to anticipate where campaign content will spread next among organic users.
Subject of Research: Computational detection of state-sponsored influence operations and prediction of user engagement on social media using temporal interaction networks.
Article Title: Temporal signatures of coordination: detecting and anticipating influence operations on social media
Article References: Mungari, S., Ferrara, E., & Luceri, L. (2026). Temporal signatures of coordination: detecting and anticipating influence operations on social media. EPJ Data Science. https://doi.org/10.1140/epjds/s13688-026-00693-8
Image Credits: AI Generated
DOI: 10.1140/epjds/s13688-026-00693-8
Keywords: influence operations, social media, coordinated behavior, graph neural networks, temporal networks, account detection, state-sponsored campaigns, user engagement prediction, computational social science, open access, Temporal, signatures
News Source: Denise Maddox. (October 29, 2026). Temporal fingerprints expose coordinated influence operations on social media. Scienmag.
