Exploring Trust-Based and Reputation-Aware Frameworks for Attack Detection in Recommender Systems A 5-Year Survey

Digital platforms use recommender systems to increase user engagement by personalizing
content for users. Because of this, they are often abused by outside influences that will ma
nipulate the system to favour themselves– as in the case of profile injection or shilling attacks.
In recent years, there has been a shift to use of non-statistical means of detection, such as
relational credibility, trust propagation, temporal consistency, and coordinated behavioural pat
terns as ways to identify suspicious users. This literature review provides a critical assessment
of the progress that has been made in areas of trust-based/reputation-aware attack detec
tion for recommender systems, summarizing key methods and identifying their strengths and
weaknesses while offering some potential future directions. The review finds that hybrid mod
els using trust, reputation, temporal signals and collusion analysis provide a greater degree
of robustness than traditional rating based detectors, but also suggests that significant chal
lenges exist with regard to cold-start users, explainability/fairness, real-world deployment and the threat of AI-generated stealth attacks. The paper concludes that research into the security of recommender systems should focus on dynamic trust modelling, graph-based detection and interpretable defence mechanisms.