With the appearance of big data’s era, some problems caused in recommendation systems are needed to solve immediately. So it is very useful to design parallel recommendation algorithms. An improved parallel item-based collaborative filtering (IP_Item-basedCF) algorithm based on Hadoop is proposed in this paper. In order to consider the influence of user’s activity, a new parameter called IUF is introduced that can give the active users soft punishment. And the user’s rating is also considered in prediction model. Finally, we evaluate the performance of our approach by using two real datasets – MovieLens and Douban. The experimental results show that this new parallel algorithm outperforms the algorithms existed and has a good scalability and speedup.
展开▼
