A Stacking-Based Ensemble Movie Recommender System

Abstract

Recommender systems are information filtering systems that influence customer behavior in the digital world by suggesting relevant items to users. In this paper, we have proposed a stacking-based ensemble movie recommender system. The proposed recommender system has been tested on the MovieLens dataset obtained from the Grouplens project. We have experimented with different combinations of base learners and meta-learners to find the best base learners and meta-learner combination for the proposed recommender system. Different regression models like Linear Regressor and K-Nearest Neighbors regressor and ensemble models like Gradient Boost, AdaBoost, Random Forest and XGBoost (eXtreme Gradient Boosting) have all been tested for this purpose. The experiments conducted show that stacked models outperform stand-alone models. It is also observed that XGBoost gives excellent results when used as a base learner and also as  a meta-learner in  the  proposed  stacked  model..