e-shop user preferences via user behavior

We deal with the problem of using user behavior for business relevant analytic task processing. We describe our acquaintance with preference learning from behavior data from an e-shop. Based on our experience and problems we propose a model for collecting (java script tracking) and processing user behavior data. We present several results of offline experiments on real production data. We show that mere data on users (implicit) behavior are sufficient for improvement of prediction of user preference. As a future work we present richer data on time dependent user behavior.