"Mining Transaction Data for Personalized Supermarket Shopping Recommendation"

6/03/05: 10:30 AM, webcast
11th Floor Large Conference Room
Host: Patrick Pantel, schedule

Abstract: In this talk, I will report our experience in real-world data mining applications for personalized shopping recommendation. We found that the transaction data is very skewed and sparse that causes some problems for widely used collaborative filtering and data mining algorithms. Therefore we developed a novel probabilistic graphical model called HyPAM for personalized shopping recommendation and empirically show that HyPAM outperforms those algorithms. Then we found that since a large portion of sales is concentrated in a small number of hot seller items, collaborative filtering recommenders, including HyPAM, usually recommend hot sellers while rarely recommend cold sellers. But recommenders are supposed to provide better campaigns for cold sellers to increase sales. We cast the problem as a rare class classification problem and used a boosting algorithm to train an ensemble of SVM classifiers to predict who is most likely to buy cold sellers. Experimental results show that our Boosting-SVM algorithm can improve from a baseline approach by about twenty-five percent for cold sellers that as low as 0.7% of customers have ever purchased.

Last updated: Mon Jun 19 17:44:06 2006