[4B2-J-3] Data mining: structures and clusters

[4B2-J-3-04] Biclustering the purchase transaction data for tea drinks

Extracting interesting patterns and its experimental evaluation in the real supermarket store

Yuma Ishida¹, 〇Yukinobu Hamuro¹, Hiroaki Maruhashi¹, Naoki Katoh¹, Takeaki Uno² (1. Kwansei Gakuin University, 2. National Institute of Informatics)

Keywords:in-store experiments, datamining, graph polishing, maximal clique

The objective of this paper is to show the case study about a knowledge discovery in the consumer behavior database, and test the effectiveness of the method, biclustering, in conducting an experimental evaluation at the real supermarket store. The biclustering we used consists of two core methods, bipartite graph polishing and maximal clique enumeration. The experiment of selling the specified health drinks with non-healthy foods showed high sales gain, but cannot be confirmed in other experiments. We learnt difficulty to implement the datamining results in real world, and get insight of the reasons.