Discovering Periodic High Utility Itemsets in a Discrete Sequence

Periodic itemset mining is the task of finding all the sets of items (events or symbols) that regularly appear in a sequence. One of the most important applications is customer behavior analysis, where a periodic itemset found in a sequence of customer transactions indicates that the customer regularly buys some items together. Using this information, marketing strategies can be tailored and product recommendation can be done. However, a major limitation of traditional periodic itemset mining is that the relative importance of each item is not taken into account and that each item cannot appear more than once at each time step of the sequence. But in real life, not all items are equally important (e.g., selling a cake yields less profit than selling a computer) and a customer may buy multiple units of the same item at the same time (e.g., many cakes). To address these factors, the task of periodic frequent itemset mining was generalized as that of periodic high utility itemset mining, where the goal is to find the sets of items that not only periodically appear in a sequence but also have a high importance (e.g., yield a high profit). This chapter provides an overview of this task and presents an algorithm to solve this problem. Moreover, a variation of this task that consisting of discovering irregular high utility itemsets is also discussed. Finally, some research opportunities are listed.