A Comprehensive Survey on Utility-Based and High-Utility Pattern Mining Techniques in Recommender Systems

Recommender systems (RSs) have become indispensable in various domains to mitigate information overload by providing personalized suggestions.While traditional RSs primarily focus on accuracy (e.g., rating prediction), modern applications demand consideration of business-centric objectives such as profit, user engagement, and long-term revenue.Utility-based recommender systems aim to optimize these objectives by integrating utility measures into recommendation models.Moreover, high-utility pattern mining (HUPM) techniques have emerged as powerful tools to identify patterns that maximize user engagement or profit in largescale clickstream data.This survey presents a comprehensive review of utility-based and high-utility pattern mining approaches in RSs.We categorize existing methods, discuss underlying theoretical foundations, analyze their strengths and limitations, and outline open challenges and future research directions.The survey covers utility function design, algorithmic advances in HUPM, hybrid frameworks combining collaborative and content-based features with utility optimization, scalability considerations in big data contexts, and evaluation metrics beyond accuracy.Finally, we highlight emerging trends such as deep learning integration, fairness-aware utility modeling, and real-time recommendation under utility constraints.This survey aims to serve as a reference for researchers and practitioners seeking to develop next-generation RSs that balance accuracy, business value, and user satisfaction.