A personalized recommender system for a luxury e-commerce website. An approach based on market basket data

A recommender system applies data mining and knowledge discovery techniques to the problem of making personalized filtering of information, products or services (Sarwar et al., 2000). Recommender systems can be classified broadly into several categories depending on the information they use to recommend. Con- tent-based recommendation systems try to recommend new items similar to those a particular user has liked in the past (Lops et al., 2011). Collaborative filtering algo- rithms base their recommendations on the ratings or behaviour of other users in the system (Ekstrand et al., 2011). The present paper is a preliminary attempt to test a recommendation engine based on collaborative filtering, designed explicitly for a luxury e-commerce website from Apulia, Italy. The ratings are implicitly expressed by users’ behaviours (purchase/not purchase). These implicit ratings (one-class data) are easy to obtain, although all of the negative examples (not purchase) and missing positive examples (purchase) are mixed together and cannot be distinguished. Problems encountered when applying standard collaborative filtering algorithms can also arise due to extreme data sparsity.