Abstract
This paper presents an integrated framework for analyzing information diffusion in social networks and detecting threats in the Darknet under partial uncertainty conditions. We adapt the Susceptible-Infected-Recovered (SIR) epidemic model to simulate information spread, estimating transmission (β = 0.12 ± 0.03) and recovery (γ = 0.08 ± 0.02) rates through hierarchical clustering of VKontakte data, achieving 23% lower RMSE than the baseline methods. For Darknet analysis, we develop a neural pipeline combining transformer-based text vectorization, feedforward classification (F1-score: 0.87), and cosine similarity search (Precision@10: 0.92), outperforming keyword search by 28%. The implemented system monitors 12K Darknet pages/hour with automated threat alerts. Results demonstrate effective modeling of viral peaks (42% population coverage at 48 hours) and precise identification of illicit content despite lexical variability and anonymity constraints. This work bridges epidemiological modeling with deep learning, offering a replicable framework for cybersecurity applications in uncertain digital environments.