E-Retail Example--Modeling Techniques

The modeling techniques employed by the e-retailer are driven by the company's data mining goals:

Improved recommendations. At its simplest, this involves clustering purchase orders to determine which products are most often bought together. Customer data, and even visit records, can be added for richer results. The two-step or Kohonen network clustering techniques are suited for this type of modeling. Afterward, the clusters can be profiled using a C5.0 ruleset to determine which recommendations are most appropriate at any point during a customer's visit.

Improved site navigation. For now, the e-retailer will focus on identifying pages that are often used but require several clicks for the user to find. This entails applying a sequencing algorithm to the Web logs in order to generate the "unique paths" customers take through the Web site, and then specifically looking for sessions that have a lot of page visits without (or before) an action taken. Later, in a more in-depth analysis, clustering techniques can be used to identify different "types" of visits and visitors, and the site content can be organized and presented according to type.