Classification-based collaborative filtering using market basket data

Collaborative filtering based on voting scores has been known to be the most successful recommendation technique and has been used in a number of different applications. However, since voting scores are not easily available, similar techniques should be needed for the market basket data in the form of binary user-item matrix. We viewed this problem as a two-class classification problem and proposed a new recommendation scheme using binary logistic regression models applied to binary user-item data. We also suggested using principal components as predictor variables in these models. The proposed scheme was illustrated with a numerical experiment, where it was shown to outperform the existing one in terms of recommendation precision in a blind test.