Hybrid popularity model for solving cold-start problem in recommendation system

This research proposes a new hybrid popularity model for solving the cold-start problem in the recommendation system. A cold-start problem arises when the target user has no rating history in the system. A hybrid popularity model combines the benefit of both the user and item popularities. The item popularity model assumes that a target user is most expected to like the top-rated items. Whereas the user popularity model presumes that a target user is likely to be influenced by the top users who have given a large number of ratings. Naturally, our proposed HPop model is built in three phases: item popularity, user popularity, and hybrid popularity. The ratio of the item and user popularities are controlled by the use of α. We use the Normalized Discounted Cumulative Gain (NDCG), as well as Precision and Recall metrics to evaluate the performance of our model and its counterparts, i.e., IPop and UPop. Using a real-world MovieLens dataset, our experiments show that the employment of the user popularity model is always more beneficial than the item popularity model. HPop performs best when α = 0.9 and worst when α = 1. The NDCG average of increases from HPop to IPop and UPop are respectively 12.22% and 8.02%. The results in terms of Precision-Recall also show a similar trend to those of NDCG. Hence, we conjecture that the performances of HPop, IPop, and UPop are stable in any evaluation metrics.