Ensemble-Based Context-Aware Recommender System Using Clustering And Singular Value Decomposition

The recommendation system is facing the problem of how to overcome sparsity, scalability, and cold start issues. Numerous studies have been done to increase recommendation accuracy and address the aforementioned issues. But there is a need for improvement. Therefore, this research proposed a context-aware recommendation system that tries to improve accuracy, scalability, and sparsity issues of recommendation by integrating ensemble clustering for clustering the dataset, SVD for dimensionality reduction, and context-aware for improving the accuracy and sparsity issue. The research also will create a new similarity metric that will be used to find similarities among user-user and/or item-item so that the sparsity of the data is decreased. we evaluate our approach on MovieLens 100k and 1M real-world datasets. The findings indicate that the SVD has performed better with high-accuracy rating prediction using the reconstructed low-dimensional matrices in comparison to the KNN by splitting the sparse and high-dimensional rating matrix into two low-dimensional matrices. While the context-aware uses the item context such as movie ontology and user contexts such as age, location, and gender to improve scalability. On comparison with the state-of-the-art approach so far, we noticed that the proposed method based on (User and Item+SVD+ensemble+Context) was the best-performing technique with an average MAE score of 0.6031. F-measure, Precision, and throughput results demonstrated that the CF recommender systems' performance could be enhanced by using context awareness in conjunction with clustering and dimensionality reduction techniques. Our analysis of findings showed that recommender systems' scalability and sparsity problems can be resolved using the suggested recommendation method.