Utility Mining with Individual Item Thresholds

• Main Authors: Yu-Te Chao, 趙育德
• Other Authors: Tzung-Pei Hong
• Format: Others
• Language: en_US
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/62271243300077912737

碩士 === 國立高雄大學 === 資訊工程學系碩士班 === 101 === Utility mining has recently attracted much attention due to its wide applications. It, however, considers the items uniformly by a single minimum utility criterion, such that the significance of the items is not actually reflected. How to develop an efficient and effective utility-based framework with the consideration of different item and itemset criteria is thus a critical issue. In this thesis, we introduce multi-criteria utility mining, which allows users to specify different minimum utility thresholds to items according to the characteristics or importance of items. In addition, two different viewpoints, respectively from minimum and from maximum constraints, are also presented to decide the utility threshold of an itemset when its items have different criteria. For the minimum constraint, the downward-closure property cannot be kept in the multi-criteria utility mining process. An effective sorting strategy and a two-phase multi-criteria approach are designed to cope with the problem.For the maximum constraint, the downward-closure property exists, such that the original two-phase approach can be easily extended to find high utility itemsets under multiple item thresholds. Finally, the experimental results on several simulation datasets show the effectiveness of the two viewpoints and the performance of the proposed approaches under different parameter settings.