Discovering Purchase Association among Brands from Purchase History

Analyzing purchase history of customers enables us to discover valuable knowledge that is helpful for developing effective sales promotion. In this respect, we shall introduce a new notion, association strength, defined for purchase history which quantitatively evaluates brand loyalty and brand switching behavior among several competing brands in certain commodity categories. We shall present two examples to explain how this notion is useful by demonstrating computational results based on real sales data in a Japanese drugstore chain. One example is brand loyalty and brand switching analysis on baby diaper. Diaper users tend to evolve over their life-cycles. Using the association strength, we can measure how more (or less) often users switch one brand to the other brand compared with the case where brand switching randomly occurs according to the market share. The experimental results obtained clarifies which brands are competing with each other and which brands have high brand loyalty. The second example discovers associative purchasing patterns among three or more brands each of which belongs to distinct commodity category. We have obtained several interesting user behaviors. The proposed notion extends conventional basket analysis since market basket analysis is performed for one purchasing opportunity but associative purchasing pattern is obtained from series of purchase opportunities in purchase history of several months. Keywords— data mining, purchase history, association strength, marketing