Dual-evolution: a deep sequence learning model exploring dual-side evolutions for movie recommendation

Influenced by external environment or individual cognition, interests of a user and attractions of a movie keep evolving over times in the movie recommendation scenario. However, existing studies on movie recommendation ignored the evolutions of user interests and movie attractions, which leads to suboptimal recommendation performance. In order to improve movie recommendation, we developed a deep sequence learning model, namely Dual-Evolution, to simultaneously explore evolutions of user interests and movie attractions for movie recommendation. Specifically, Dual-Evolution first extracted temporary user interests and movie attractions by learning behavior sequences of the user and the movie. And then, Dual-Evolution modeled evolution processes of temporary user interests and movie attractions by considering not only their sequential association but also their relative importance. Further, we conducted experiments for two tasks on two benchmark real-world datasets. Experimental results indicate that Dual-Evolution significantly outperforms mainstream movie recommendation methods in the movie rating prediction as well as the top-N recommendation, which inspires online movie platforms more reasonably designs movie recommender systems.

Automated summary

This study addresses the issue of neglecting the evolutions of user interests and movie attractions in movie recommendation, leading to suboptimal performance. To improve movie recommendation, they propose a deep sequence learning model, Dual-Evolution, which explores the evolutions of user interests and movie attractions simultaneously.