A hybrid scalable collaborative filtering based recommendation system using ontology and incremental SVD algorithm

Abstract

Recommendation system is a dynamic information filtering system based on machine learning algorithm. Recommendation engine filters large amounts of information generated through the interaction of users to web-portals generate the recommendation to fulfill the expectation of the online users. Collaborative filtering is most popular and efficient method to generate the proper recommendation. Current method of CF is unable to handle the most popular problem of recommendation like data sparsity, scalability, recommendation inaccuracy and big error in predication. To handle the such problem of recommendation this research develops a new hybrid CF approach to addresses the problem of CF like sparsity, scalability inefficiency in recommendation improve the accuracy prediction using the incremental approach of singular value decomposition and ontology approach. To enhance the prediction and recommendation accuracy ontology based approach is applied. This approach is decomposed into two parts offline phase and online phase. In offline phase of collaborative filtering method, use dimension reduction technique like singular value decomposition (SVD) and also use the clustering method to form the clusters of most similar users and items depends upon the preferences which significantly improve the scalability of collaborative filtering. In online phase of recommendation the incremental SVD is used to find out the most relevant item to the new item. The performance of the recommendation system is evaluated through evolution parameter likes Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) over the two real word dataset like Movielens and Flixster Dataset. The result shows that the implemented method effective is to solve the problem of user scalability, data sparsity and improve the performance of prediction and recommendation.