Basket analysis for graph structured data

Abstract

The Basket Analysis derives frequent itemsets and association rules having support and confidence levels greater than their thresholds from massive transaction data. Though some recent research tries to discover wider classes of knowledge on the regularities contained in the data, the regularities in form of the graph structure has not been explored in the field of the Basket Analysis. The work reported in this paper proposes a new method to mine frequent graph structure appearing in the massive amount of transactions. A specific procedure to preprocess graph structured transactions is introduced to enable the application of the Basket Analysis to extract frequently appearing graph patterns. The basic performance of our proposing approach has been evaluated by a set of graph structured transactions generated by an artificial simulation. Moreover, its practicality has been confirmed through the appliaction to discover popular browsing patterns of clients in WWW URL network.