GPU-based Collaborative Filtering Recommendation System using Task parallelism approach

Abstract

Collaborative filtering is one among the top most preferred techniques when implementing recommendation systems. In recent times, more interest has turned towards parallel GPU-based implementation of collaborative filtering algorithms. Concurrent way of solving any problem is more preferable by everyone nowadays. The objective of GPU-based collaborative filtering recommender system is to produce recommendations in parallel and choosing the best among all. We have proposed three different methods namely Parallel Item Average Computation (PIAC), Parallel User Based Collaborative Filtering (PUBCF) and Parallel Item Based Collaborative Filtering (PIBCF).We have evaluated all these methods with standard evaluation metrics. As a result of task parallelism, the PIBCF method produces optimum choice for providing better recommendation results.