Supporting retrieval of configurable goods based on statistical analysis of combination constraints

Configurable goods are becoming a popular style for e-commerce web shopping sites in which buyers can configure a product of their needs from menus listing components. In this paper, we propose a sophisticated system support for designing web menus for configurable goods. We discuss evaluating the correlations between component classes of configurable goods. Such correlations can be used to design web menus which cause fewer trial errors and give an aggregated view of product constraints. Choosing a proper quantitative measure for correlation is an important issue here. We compare a number of statistical and mining methods by experiments and show that Cramer's coefficient is most suitable for this problem. Then we present an algorithm which generates a tree structure for web menus such that closely correlated component classes are clustered, and hence users can easily select components.