Poisoning and shilling attacks remain a serious threat to recommender systems, especially as attackers increasingly mimic plausible profile statistics. This paper proposes an architecture-independent behavioral detection layer that models user interactions as short-window phase-dynamic trajectories rather than static aggregates. Interaction logs are transformed into temporal signals, reconstructed in phase space by delay embedding, and summarized by a compact 15-dimensional portrait combining recurrence-based, entropy-based, spectral, and stabilizing statistical descriptors. In a controlled targeted injection protocol evaluated over 10 independent runs, the statistical baseline achieved PR-AUC = 0.723 ± 0.037 and TPR@1%FPR = 0.029 ± 0.006, the dynamic block achieved PR-AUC = 0.831 ± 0.011 and TPR@1%FPR = 0.220 ± 0.050, and the full portrait achieved PR-AUC = 0.872 ± 0.017 and TPR@1%FPR = 0.291 ± 0.043. Sensitivity analysis showed that recurrence-only descriptors were parameter-sensitive, whereas the extended dynamic block formed a stable high-performance region across a broad range of embedding settings. An IQR-normalized aggregated risk score further demonstrated clear post-window regime separation during injection periods. The results indicate that poisoning attacks primarily deform the temporal organization of behavior rather than only first-order statistics. The proposed phase-dynamic portrait is therefore best interpreted as a complementary behavioral risk-scoring layer for auditing, filtering, and monitoring rather than as a standalone defense.

This paper presents a probabilistic-temporal modeling approach for analyzing data processing stages in a personalized recommendation system for digital heritage collections. The methodology is based on (Graphical Evaluation and Review Technique) GERT network formalism, which enables the representation of complex probabilistic workflows with feedbacks and alternative branches. For each processing stage, corresponding GERT-schemes were developed, and equivalent transfer functions were derived. Using Laplace transform inversion techniques, probability density functions of processing time were recovered, followed by the calculation of key statistical metrics, including expectation, standard deviation, and quantiles. The results demonstrate that the proposed approach allows for detailed temporal performance evaluation, including the estimation of time exceedance probabilities at each stage. This provides a quantitative basis for optimizing recommendation system design and highlights the applicability of GERT-based modeling to intelligent data-driven services in the cultural domain.