RECENT ADVANCES IN UTILITY-DRIVEN PATTERN MINING

Abstract

Knowledge discovery is a key part of Artificial Intelligence that focuses on finding valuable, hidden patterns and insights in large amounts of complex data. Over the last two decades, Utility-Driven Pattern Mining (UDPM) has emerged as a powerful unsupervised knowledge discovery paradigm to address a critical business need: identifying patterns that are highly valuable, not just frequent. UDPM has undergone revolutionary diversification in algorithmic approaches and utility quantification frameworks, enabling unprecedented capture of economic significance beyond traditional frequency-based models. In this paper, we highlight the recent advances and innovative algorithmic evolution in UDPM, ranging from early candidate-generation strategies to advanced one-phase and memory-efficient methods. Furthermore, we describe its broad theoretical extensions that have been developed to address practical challenges such as negative utilities, temporal aspects, recurrence, and dynamic environments, providing researchers and practitioners with a clear understanding of what has happened and what is happening in this rapidly advancing field.