Auditing Representation-Space Fraud Detection: A Bayesian IRL Framework for Calibrated Uncertainty and Its Limits
Shivam Swarup, Divya Prakash Shrivastava, Rakesh Thakur
Abstract
Shilling detectors for recommender systems are almost always evaluated by one number, detection AUROC, with no account of when it can be trusted. We audit representation-space detection with VEIL (Variational Evaluation via Inverse-reinforcement Learning), a pairwise (Bradley–Terry) Bayesian reward-inference framework over frozen encoder embeddings, after the Bayesian auditing paradigm for LLM alignment [1]: (A) a base posterior; (B) a hierarchical ARD extension with a learned confidence; and (C) a label-routed organic/attacker mixture used as an evidence-accumulation check. We prove (Prop. 1) that no detector built on a representation invariant to some transformation of user profiles can beat chance against an attack indistinguishable from organic profiles under that invariance. Detection is near-perfect against synthetic-structure attacks (AUROC 0.936–1.000) but collapses under donor-mimicry, which clones a real user’s item set and splices in 1–3 target items: to chance on MovieLens-1M (0.48–0.51), partially on Amazon Musical Instruments, and least on CiaoDVD. Donor-mimicry is
not undetectable: a non-learned baseline, each profile’s maximum raw item-set Jaccard similarity to any other, reaches AUROC 0.92–0.97 (average precision 0.50–0.57) on all four donor-mimicry configurations, including ML-1M, where every embedding-based score is at chance. The signal is
in the raw profiles; LightGCN’s representation discards it, placing injected profiles closer to their true donor than its zero-propagation MF reduction does (distance 1.98 vs. 3.09). Severity varies by dataset, encoder, and detection method, with no one axis predicting the others. The limit
lies in the representation, not the attack, and raw-similarity checks are a necessary complement to any learned detector