A NOVEL RECOMMENDER SYSTEM USING DEEP LEARNING TO ENHANCE THE RATING PREDICTION AND TOP-N RECOMMENDATIONS

Abstract

Different Recommender System Algorithms such as Content-Based and Collaborative Based have been developed by researchers and data scientists in order to filter a large amount of information available on the internet and hence, recommend only the relevant and important content based on the personalized interests of users. Information acquired explicitly by collecting users’ ratings for an item lead to the problem of data sparsity. Many researchers have been working towards the improvement of rating prediction accuracy by integrating the auxiliary information along with the ratings provided by the users. It has been observed in related works that integrating the textual data along with rating data has brought an improvement in the accuracy of estimating the score given to an item by a user and the ranking of top-n recommendations. However, document modeling approaches are different in different research papers. This Project proposes a unique deep neural network text analysis model that includes newly discovered neural network architecture, Capsule Networks stacked on bi-directional Recurrent Neural Network (Bi-RNN) for developing a robust representation of textual descriptions of items and users. The Deep Neural Network text analysis model is integrated with the Probabilistic Matrix Factorization to generate the recommendations. The proposed Model is called as “CapsMF” since it applies the advanced neural network architecture Capsule Networks (Caps) for document representation and MF represents Matrix factorization that is being enhanced to improve recommendations. The experiment is performed on two real amazon datasets and has shown that the rating prediction accuracy and the recall, as well as the precision of top-n recommendations, have vi

improved in comparison to the basic and hybrid Recommendation System Algorithms. Also, Capsule Networks stacked with Recurrent Neural Networks (RNNs) have outperformed the baseline models that involve single Convolutional Neural Networks (CNN) or CNN combined with Bi-RNN. We have also compared different deep learning algorithms and have shown how different text representations affect the recommendations accuracy.