Analysis on merchandise hierarchy via clustering retail records

Merchandise hierarchy plays an important role in modern retail business. In this paper, we suggest a quantitative consumer-related evaluation method of merchandise hierarchy via clustering the retail records. The retail records contains much information reflecting the consumer buying behavior, and it should be effectively utilized to judge whether a predefined merchandise hierarchy is appropriate for a specific retailer´s business. Here, we mainly mine the complementary information between products from the retail records, that reflect some crucial consumer buying habits and can be used to evaluate predefined merchandise hierarchy. Concretely, the spectral clustering algorithm is adopted to obtain the cluster assignments of items on each level of merchandise hierarchy and Normalized Mutual Information is used to compare the cluster results and the corresponding merchandise hierarchy. We conduct some experiments on a real supermarket retail records and get some interesting and valuable consumer insights. Besides the merchandise hierarchy evaluation work, we further provide a preliminary scheme that can refine the merchandise hierarchy by clustering the retail records.