Systematic literature review on enhancing recommendation system by eliminating data sparsity

The aim of this project is to develop an approach using machine learning and matrix factorization to improve recommendation system. Nowadays, recommendation system has become an important part of our lives. It has helped us to make our decision-making process easier and faster as it could recommend us products that are similar with our taste. These systems can be seen everywhere such as online shopping or browsing through film catalogues. Unfortunately, the system still has its weakness where it faced difficulty in recommending products if there are insufficient reviews left by the users on products. It is difficult for the system to recommend said products because it is difficult to pinpoint what kind of users would be interested in the products. Research studies have used matrix factorization as the standard to solve this issue but lately, machine learning has come up as a good alternative to solve data sparsity. This project compares results of the recommendation system using RMSE to see how each proposed methods performs using three different datasets from MovieLens. We have selected two models – matrix factorization with SVD and deep learningbased model to evaluate these approaches and understand why they are popular solution to data sparsity. We have found that SVD brought in a lower RMSE as compared to deep learning. The reason behind this was discussed in the latter chapter of this thesis. We have also found possible research in capitalising categorical variables in recommendation system and the experiment achieved a lower RMSE score as compared to SVD and deep learning, showing the many possibilities of the future directions of the research in recommendation system