Service recommendation using conditional restricted Boltzmann machines

We propose methods based on the conditional restricted Boltzmann machine (CRBM) for the service recommendation. First, we construct a CRBM model, the individualised characteristics of customers and indexes of satisfaction have been encoded into its conditional units, and the using status of services has been encoded into its visible units. Next, a method for dynamically adjusting learning rates is proposed to improve the training process of the CRBM. Finally, we develop a neighbourhood-based approach to further boost recommendation results. The evaluation on a dataset extracted from a manufacturing company, validates that the above-proposed methods have highly practical relevance to the service recommendation problem in real world business.