Ekle, Ocheme Anthony; Eberle, William; Christopher, Jared. “Adaptive DecayRank: Real-Time Anomaly Detection in Dynamic Graphs with Bayesian PageRank Updates.” Applied Sciences (Switzerland) 15, no. 6 (2025): 3360. https://doi.org/10.3390/app15063360.
Detecting unusual behavior in large, constantly changing networks—like catching hackers in action, spotting fake news, or identifying suspicious bank transactions—is extremely important. But many current tools that analyze these networks are based on fixed snapshots, which makes them less effective when things change quickly.
This study introduces Adaptive-DecayRank, a new method that can spot strange or unexpected patterns in real time, even as the network evolves. It builds on a well-known technique called PageRank (which Google uses to rank websites) but makes it smarter by letting each part of the network adjust how quickly it “forgets” past activity, based on what’s happening now. This helps the system quickly recognize sudden changes in the structure of the network—like a burst of unusual connections or suspicious activity.
The researchers tested this new method on several real-world cybersecurity datasets, including ones from the U.S. government and university research projects, and also on complex simulated networks. The results showed that Adaptive-DecayRank was significantly better at detecting anomalies compared to other leading tools, catching more threats with greater accuracy—even in fast-changing environments.
Figure 1. An illustraction of graph representation: (a) Static graph, G, and (b) evolving dynamic graph, 𝒢=(𝑉𝑡,𝐸𝑡,𝒯), showing a series of graph snapshots with edge insertions, deletions, and node insertions over time.