Modeling Temporal Behavior of Awards Effect on Viewership of Movies

Abstract

The “rich get richer” effect is well-known in recommendation system. Popular items are recommended more, then purchased more, resulting in becoming even more popular over time. For example, we observe in Netflix data that awarded movies are more popular than non-awarded movies. Unlike other work focusing on making fair/neutralized recommendation, in this paper, we target on modeling the effect of awards on the viewership of movies. The main challenge of building such a model is that the effect on popularity changes over time with different intensity from movie to movie. Our proposed approach explicitly models the award effects for each movie and enables the recommendation system to provide a better ranked list of recommended movies. The results of an extensive empirical validation on Netflix and MovieLens data demonstrate the effectiveness of our model.