Learning Customer Profiles Using Unlabelled Data

E-commerce sites often recommend products they believe a customer is interested in buying. Many web sites have started to embody recommender systems as a way of personalizing their content for users. This paper presents a recommender system that exploits supervised learning methods to learn user profiles from items previously rated by users. Profiles are used to find, classify, or rank items that are likely to be of interest to the user. A major concern with supervised learning techniques is that they often require a large number of labelled examples to learn accurately. Our proposal to reduce the amount of labelled data required is an algorithm that can learn effectively from a small number of labelled examples augmented with a large number of unlabelled examples. Experiments on a real dataset show that the proposed method is effective.