User-centered requirements elicitation for explainable AI transparency in recommendation and advertising systems

Abstract

Modern recommendation and advertising systems increasingly rely on complex artificial intelligence (AI) models whose opaque decision-making processes limit user understanding and trust. While explainable artificial intelligence (XAI) techniques aim to improve transparency, excessive disclosure can increase privacy concerns and psychological discomfort. This study addresses the lack of structured approaches for regulating transparency in user-facing AI systems. We propose the optimal transparency and psychological safety (OTPS) framework, which regulates explanation depth, timing, and user control to balance interpretability with psychological safety. The framework is implemented through a modular architecture consisting of an explanation generation module, transparency controller, and user interface layer. A user survey involving 35 participants was conducted to evaluate perceptions of transparency, trust, and psychological comfort. Statistical analysis, including reliability testing and response distribution evaluation, indicates strong user preference for adaptive transparency mechanisms. The results demonstrate that regulated transparency improves user trust and usability without introducing significant system overhead, providing practical design guidance for explainable AI systems in recommendation and advertising platforms.