Collaborative filtering using probabilistic matrix factorization and a Bayesian nonparametric model
Nurudeen Sherif, Gongxuan Zhang · 2017
In recent times, numerous web-based systems that offer various kinds of services ranging from e-commerce to video and audio streaming are overwhelmed with influx of data. These systems have been introducing various recommender algorithms into their platforms to filter information from their voluminous databases. These recommendation engines help users to save time by providing them with results that suite their taste. Researchers and software engineers have developed numerous algorithms to recommend items more accurately however, very little research have been done on the scope of music recommendation using social data. Orthodox methods of recommendation utilized ratings supplied by users to each item as the basis of recommendation. The accuracy of prediction is significantly affected when ratings are not provided for some items or when the user is new to the platform and has not rated any item yet. It is also difficult to recommend new items that have not yet being rated yet. This problem is termed “cold start”. In order to eliminate this problem we propose the Collaborative Filtering (CF) using Probabilistic Matrix Factorization (PMF) and a Bayesian Nonparametric Model (BNM). A comprehensive experiment on music and social datasets from different source indicates a higher accuracy of prediction and performance.