Personalized Recommendation Tool Learning via Autonomous Language Agents

Collection of recent IR and recommender-systems papers exposing several security-relevant themes: LLM-based agent architectures (PRTA) that orchestrate external recommendation models/tools raise risks of prompt-injection, unauthorized tool invocation, and leakage of user profiles or candidate lists. Platforms that unify or visualize retrieval workflows (GraphContainer, Graph RAG tooling) can increase attack surface and accidental disclosure when applied to sensitive knowledge graphs. Multiple papers release code, datasets, and benchmarks (UniRank, DeltaGate, CAPTS, occupation-coding repo), and