Content based recommender systems have the drawback of recommending only similar items to a user's particular taste, irrespective of the item's popularity. Collaborative Filtering based systems face the problem of data sparsity and expensive parameter training. In this paper, a combination of content-based, model and memory-based collaborative filtering techniques is used in order to remove these drawbacks and to present predicted ratings more accurately. The training of the data is done using feedforward backpropagation neural network and the system performance is analyzed under various circumstances like number of users, their ratings and system model. © 2014 IEEE.