Improving efficiency of recommender systems

By learning from the past behaviors of user transaction records, recommender systems can help people to nd interesting products from many other products. In a collaborative ltering based recommender system, products are regarded as features. However, there are usually quite a lot of products to be considered. A recommender system would be very inefficient if such a large number of products are processed before making any recommendations. We propose a method which applies a self-constructing clustering technique to reduce the dimensionality related to the number of products. As a result, the processing time for making recommendations is much reduced without degrading the accuracy of recommendations.