At present online marketplaces are flooded with numerous users having diversified choices across profusely present products. Recommender systems were developed to suggest items, friends or movies etc. to online users based on their profiles. Memory based Collaborative Filtering (CF) is one of the widely used approaches to build recommender systems. Despite quite successful, memory based CF approaches fail on scalability; which is the deteriorated performance of these approaches on the influx of new users/items into the system. This study presents Biclustering based Collaborative Filtering (BBCF)- an augmentation to existing recommender approaches and the comparison of the performance of the proposed approach on two datasets; Movielens100K and Jester, belonging to different domains and have different volumes, density, and user to item ratio. The results presented in this paper demonstrate the outperformance of BBCF systems compared to the state-of-the-art rating prediction approaches. One of the interesting findings based on the empirical results is that the performance of BBCF is better than the baseline approaches in terms of MAE, Recall, Item Coverage and Throughput. In addition, this study presents a comprehensive survey of biclustering approaches used in Collaborative Filtering systems, the impact of the number of biclusters and the overlapping degree on prediction and recommendation quality along with the limitations and some interesting research avenues in BBCF systems. (C) 2018 Elsevier Ltd. All rights reserved.
展开▼
