Implementation of Regularized Singular Value Decomposition in Collaborative Filtering Book Recommendation System

Abstract

At the school level, time is limited by the system of lesson hours. This makes students have to use their time wisely before changing lesson. However, choosing appropriate reading material often requires more time which results in wasted class hours. The development of a recommendation system using the Collaborative Filtering (CF) and Regularized Singular Value Decomposition (SVD) methods was chosen to solve the problem of students having difficulty finding books in the library. The data used is student interaction data with books in the form of ratings which are collected directly and processed to provide recommendations. The results of applying SVD in predicting ratings and looking for appropriate latent features to describe the characteristics of students and books produce MAE and RMSE values of 0.478 and 0.686. The research conducted also shows that the appropriate number of latent factors or features and the addition of regularization have an effect on increasing prediction accuracy. The predicted value of the rating is then used to provide personal book recommendations and the latent feature values of the books found are used in calculating cosine similarity to provide non-personal recommendations.