Top-N Recommendation using Bi-Level Collaborative Filtering
Suman Banerjee, Pratik Banjare, Mamata Jenamani, Dilip Kumar Pratihar · 2017
Recommendation systems have been developed to provide personalized items to user based on his/her preferences. User-based collaborative filtering has been the most successful recommendation system, providing the most reasonable level of accuracy. However, with the continuous rise in number of users over the Internet, the algorithm suffers from the scalability problem. To cope up with this issue, item-based collaborative filtering (CF) system has been developed, which is scalable, however, not as accurate as user-based recommendation algorithms. In item-based CF similarity is computed among the entire set of items every time, which is not much reasonable as the entire set of items contains many items which user have no interest upon. In this paper, we have proposed a methodology for providing top- N recommendation, which is basically an improved version of item-based CF. The proposed methodology consists of mainly two parts. In the first part, we reduce the number of items to work upon by sorting out items that is more likely to be fit in user's preference. Secondly, we compute the similarity among the user rated items and candidate recommendable item, generated in the first part. Finally, the proposed methodology has been implemented with movielense data set. Reported results in this paper shows that the proposed methodology improves the recommendation accuracy with less computational time.