Mining High Utility Itemsets with Regular Occurrence
Résumé
High utility itemset mining (HUIM) plays an important role in the data mining community and in a wide range of applications. For example, in retail business it is used for finding sets of sold products that give high profit, low cost, etc. These itemsets can help improve marketing strategies, make promotions/advertisements, etc. However, since HUIM only considers utility values of items/itemsets, it may not be sufficient to observe product-buying behavior of customers such as information related to "regular purchases of sets of products having a high profit margin". To address this issue, the occurrence behavior of itemsets (in the term of regularity) simultaneously with their utility values was investigated. Then, the problem of mining high utility itemsets with regular occurrence (MHUIR) to find sets of co-occurrence items with high utility values and regular occurrence in a database was considered. An efficient single-pass algorithm, called MHUIRA, was introduced. A new modified utility-list structure, called NUL, was designed to efficiently maintain utility values and occurrence information and to increase the efficiency of computing the utility of itemsets. Experimental studies on real and synthetic datasets and complexity analyses are provided to show the efficiency of MHUIRA combined with NUL in terms of time and space usage for mining interesting itemsets based on regularity and utility constraints.