A novel matrix decomposition technique in collaborative based re-commender systems

Recommender systems are systems that aim to accurately predict the evaluations that users will give to any item in advance. Collaborative filtering method, which is one of the techniques used in recommender systems, takes into account the common characteristics of users and items. In collaborative filtering based recommender systems, matrix decomposition is one of the most commonly used techniques. In matrix decomposition, Singular Value Decomposition (SVD) and Non-Negative Matrix Factorisation (NMF) based approaches are widely used. Although these methods are very good at dealing with the scalability problem, their complexity is high. In this thesis, the Truncated-ULV decomposition (T-ULV) technique is proposed as an alternative technique to improve the accuracy and quality of the recommendations. The proposed method is tested with widely used publicly available datasets. Standard metrics were used to evaluate the performance of the model. In all experiments, T-ULV outperformed the existing matrix decomposition techniques. Moreover, the proposed method solves the problems of cold start and sparsity reduction.