DYNAMIC CLUSTERING BASED ON MAXIMUM FREQUENT ITEMSETS FOR ANALYSING PURCHASE PATTERN

摘要

A novel dynamic clustering approach, dynamic clustering based on the maximum frequent itemsets (MDCA), is proposed to find the clusters interesting for purchase pattern analysis. The algorithm consists of two stages: initial clustering and optimum clustering. In initial clustering, maximum frequent itemsets are introduced to form initial clusters from the itemset attributes. To cope with the problem of dynamically changing purchase patterns, concept lattices are employed in our algorithm to calculate dynamically the maximum frequent itemsets. The optimum clustering is based on the result of the initial clustering where a similarity criterion is proposed to support the combination of concept and itemset attributes. The criterion integrates semantic distance and link, which are used to implemented to evaluate the distance between concept attributes, and to measure the closeness of maximum frequent itemsets, respectively. With the proposed MDCA approach initial clusters are generated dynamically with maximum frequent itemset and concept lattice, and the similarity criterion is introduced to implement the optimum clustering. Our experiments based on the purchase records of a big supermarket demonstrate the efficiency of the approach. The algorithm is characterised by short processing times and low information loss.