Mitigating Sparsity and Cold Start Problem in Collaborative Filtering using Cross-domain Similarity

Collaborative filtering (CF) has proven to be the most prominent and preferable approach in the recommendation systems due to its simplicity and effectiveness. The fundamental philosophy of CF-based recommendation is that similar users have the same rating patterns and similar items obtain similar ratings. However, the accuracy of CF significantly degrades when the dataset becomes highly sparse due to the unavailability of sufficient rating information of a user. The existing similarity measures have not been able to cope up this problem that very well, and this has been the motivation of this paper, in which we aim to improve the performance of collaborative filtering by utilizing the rating information of common users who belong to multiple domains. This approach can albeit both the sparsity (inadequate rating information) and the cold start (no rating information at all) problems. The effectiveness of the proposed approach is examined with the Amazon dataset and the experimental results show that the proposed approach attains high accuracy than the existing CF-based recommendation approaches.