A novel recommendation system comprising WNMF with graph-based static and temporal similarity estimators

Users’ similarity plays a crucial role in the Collaborative Filtering (CF)-based Recommendation Systems (RS). The CF uses a user-item matrix to estimate this similarity. However, the user-item matrix-based similarity performs poorly during cold-start and temporal variations in users’ behavior. To overcome this limitation in this paper, a combined static and temporal similarity- based CF is presented, which estimates the final ratings using a weighted sum of four independent ratings. The first rating is estimated through the similarity estimated using the Affinity Propagation (AP) applied to a reduced rank coefficient matrix generated by factorizing the user-item rating matrix using Weighted Nonnegative Matrix Factorization (WNMF). The remaining three ratings are estimated by employing graph-based similarity measures on bipartite graphs made between the user and their static (age and occupation), and temporal (purchasing pattern) characteristics. Finally, the performance of the proposed algorithm is evaluated using the MovieLens1M and Ta-Feng grocery datasets under different scenarios. The results show that the proposed approach not only outperforms the traditional techniques but also outperforms the Google Wide & Deep and CoupledCF models by 22.52, 2.06, and 11.26, 2.75 % for Hit Ratio (HR) and Normalized Discounted Cumulative Gain (NDCG), respectively.