Predicting purchase preferences using semi-supervised one-class SVM with graph kernels
Yasutoshi Yajima · 2007
This paper provides a method for predicting purchase preferences of customers for a specific target store based on their past purchase transactions as well as on their demographic information. We use a kernel-based semi-supervised learning approach in which Laplacian kernel matrices are exploited. Unlike the conventional kernel-based approaches such as support vector machines, the proposed method makes predictions by solving a system of sparse linear equations. We demonstrate that the proposed method can be applied to real world data sets with a huge number of customers very efficiently, and that the accuracy of the method is reasonably high compared with the conventional decision tree approaches.