A Survey of High Utility Pattern Mining Algorithms for Big Data

Abstract

High utility pattern mining is an essential data mining task with a goal of extracting knowledge in the form of patterns. A pattern is called a high utility pattern if its utility, defined based on a domain objective, is no less than a minimum utility threshold. Several high utility pattern mining algorithms have been proposed in the last decade, yet most do not scale to the type of data we are nowadays dealing with, the so-called big data. This chapter aims to provide a comprehensive overview and a big-picture to readers of high utility pattern mining in big data. We first review the problem of high utility pattern mining and related technologies, such as Apache Spark, Apache Hadoop, and parallel and distributed processing. Then, we review recent advances in parallel and scalable high utility pattern mining, analyzing them through the big data point of view and indicate challenges to design parallel high utility pattern mining algorithms. In particular, we study two common types of high utility patterns, i.e., high utility itemsets (HUIs) and high utility sequential patterns (HUSPs). The chapter is concluded with a discussion of open problems and future directions.