Temporal fingerprints for identity matching across fully encrypted domains

Shahar Somin, Keeley Erhardt, Tom Cohen, Jeremy Kepner, Alex PentlandNature Communications 2025 · DOI

TL;DR

Researchers have developed a way to link pseudonymous profiles across different encrypted platforms—like Telegram, Twitter, and crypto markets—using only the timing of their actions. By analyzing "bursty" patterns of activity (the specific gaps between posts or trades), this framework creates a unique "temporal fingerprint" that remains stable for over a year and is remarkably resilient to missing data.

What Problem Does This Paper Address?

In the age of generative AI and end-to-end encryption, identifying coordinated inauthentic behavior (like bot farms or fraudulent trading rings) has become a game of cat and mouse. Traditionally, investigators relied on content analysis (what people say) or structural analysis (who they follow).

However, LLMs can now generate infinite variations of a narrative, making content-based detection nearly impossible. Meanwhile, encryption hides private attributes, and social structures vary so wildly between platforms (e.g., a trading network looks nothing like a social media follow-graph) that linking them is a massive technical hurdle. The authors sought a "content-agnostic" signal that is intrinsic to the human (or bot) behind the screen, regardless of which platform they are using.

Methods

The core of the paper is the inter-event bursty model. Instead of looking at what an account does, it looks at the time elapsed between consecutive actions.

  1. Inter-event Time (IET): For every profile, the researchers calculate the sequence of time gaps between activities.
  2. Kolmogorov-Smirnov (KS) Statistic: To compare two profiles, they use the KS test—a non-parametric method to determine if two sets of data come from the same underlying distribution. If the "rhythm" of Profile A’s activity matches Profile B’s, they are flagged as a potential match.
  3. Bursty-TGNN: To handle massive datasets (like 250,000 traders), the authors introduced a Temporal Graph Neural Network. This model treats profiles as nodes in a "similarity network" where edges represent temporal closeness, allowing the system to learn complex, latent embeddings for identity matching at scale.

Datasets

The study validated the model across several high-stakes environments:

  • Ethereum Financial Markets: Data from 500 different crypto-token marketplaces, involving over 250,000 daily traders. Ground truth was established using unique wallet addresses.
  • Cross-Platform Social Media: A manually labeled dataset of 266 profiles across Twitter (X), Telegram, and Instagram.
  • Reddit & Telegram: Activity across 2,848 sub-reddits and 175 Telegram channels to test intra-platform, cross-domain matching.

Key Insights and Ideas

  • The "Network of Networks" Theory: The authors argue that seemingly disconnected platforms are actually linked by individuals acting as "bridges." When an external event (a "shock wave") occurs, these individuals react across multiple platforms, creating synchronized temporal bursts that reveal their shared identity.
  • Temporal Stability: Surprisingly, these fingerprints are not fleeting. The study found that a user’s temporal "rhythm" remains recognizable even after a 52-week delay.
  • Resilience to Data Omission: The model is robust. Even if 90% of a profile's activity data is missing or hidden, the remaining 10% often contains enough "bursty" signal to maintain a high level of identification accuracy.
  • Time > Structure: While a person might have different friends on LinkedIn than on Telegram, their internal "biological or behavioral clock" for when they check their phone or execute a trade remains consistent.

Key Results

  • High Accuracy: On financial markets, the model achieved an average AUC of 0.86, significantly outperforming state-of-the-art structural and temporal baselines.
  • Long-term Matching: The system correctly identified 35% of profiles after a full year by looking at only the top 10 candidates.
  • Scalability: In the massive 500-marketplace test, the Bursty-TGNN maintained an AUC of 0.78 and achieved 96% precision for the top-100 predicted matches.
  • Social Media Success: The model achieved an AUC of 0.89 for matching identities across different Telegram channels and 0.77 across different sub-reddits.

Limitations and Future Work

The primary limitation is the data volume requirement. To build a reliable distribution of inter-event times, a profile must have a minimum threshold of activity; a "lurker" who posts once a month cannot be easily fingerprinted.

The authors suggest several paths for future research:

  • Similarity Measures: Exploring alternatives to the KS statistic that might be more computationally efficient for real-time deployment.
  • Supervised Fine-tuning: While the current model is largely unsupervised, incorporating a small set of labeled "anchor" identities could further boost precision in low-resource settings.
  • Adversarial Tactics: Investigating how sophisticated actors might attempt to "obfuscate" their temporal fingerprints by intentionally delaying actions or using jitter algorithms.

Summary updated

Abstract

In the digital age, coordinated inauthentic behavior threatens societal stability, markets, and security. Advances in generative AI amplify these threats, enabling effortless content creation, amplifying actors' influence. Detection is hindered by cross-domain activity, where pseudonymous profiles operate across encrypted platforms, and by privacy constraints limiting content analysis. In this study, we propose a robust and scalable cross-domain identity matching framework, based on bursty dynamics, independent of content or interaction data. It outperforms state-of-the-art temporal and structural approaches, remains resilient to incomplete data, and correctly identifies 35% of profiles after 52 weeks. It scales effectively, attaining AUC 0.78 when matching identities across 500 marketplaces with over 250k daily traders. By framing identity matching within the "network of networks" perspective, we demonstrate how coordinated behavior propagates across domains. This dual methodological and theoretical contribution paves the way for innovative strategies to combat digital threats in an increasingly complex and adversarial landscape.
BibTeX
@inproceedings{somin2025temporal,
  author = {Shahar Somin and Keeley Erhardt and Tom Cohen and Jeremy Kepner and Alex Pentland},
  title = {{Temporal fingerprints for identity matching across fully encrypted domains}},
  booktitle = {Nature Communications},
  year = {2025},
  doi = {10.1038/s41467-025-64785-1},
}
Figure