Trust-Based Context-Aware Collaborative Filtering Using Denoising Autoencoder

Abstract

In recent times, extensive studies have been initiated to leverage deep learning strategies to enhance context-aware recommendation. Classical collaborative filtering approaches have shown potency in a wide variety of recommendation activities; however, they are inadequate to grasp dynamic interactions between people and products in addition to data sparsity and cold start problem. There is indeed a burst of attention in using deep learning to recommendation systems owing to its nonlinear modeling potential. In this article, we implement the idea of denoising autoencoders for personalized context-aware recommendation. In specific, the proposed method comprises of split item rating according to all contextual conditions resulting in fictive items that is being fed into the denoising autoencoder augmented with trust information to overcome sparsity, referred to as Item Splitting_Trust-based collaborative filtering using denoising autoencoder (IS_TDAE). Thereby, IS_TDAE is able to predict context-based item preference under all possibility of context situation and able to suggest recommendation according current context situation of the target user. Experiments conducted on two public datasets demonstrate the effectiveness of the proposed model significantly outperforms in top-N recommendation task over the state-of-the-art recommenders.