In this paper, we propose a deep recommender system that utilizes purchase history data of customers when ratings for products are not available at all. The purchase history data of these customers are obtained between May 2014 and July 2015 from nationwide branches of a particular supermarket chain. If the customer’s purchase cycle for a particular product is shorter than the average purchase cycle, then the preference for the product will be high. Using this idea, we generate rating data from these customers’ purchase history data. Among collaborative filtering recommender systems using rating data, we know that the restricted Boltzmann machine (RBM) shows the best performance in terms of prediction accuracy. Thus, for recommender system wepropose the use of conditional RBM, which utilizes additional information on whether or not to purchase. Through an experiment we show that conditional RBM works best in predicting the ratings for products. This recommender system is expected to be useful in future.