Combining multiple metadata types in movies recommendation using ensemble algorithms

Abstract

In this paper, we analyze the application of ensemble algorithms to improve the ranking recommendation problem with multiple metadata. We propose three generic ensemble strategies that do not require modification of the recommender algorithm. They combine predictions from a recommender trained with distinct metadata into a unified rank of recommended items. The proposed strategies are Most Pleasure, Best of All and Genetic Algorithm Weighting. The evaluation using the HetRec 2011 MovieLens 2k dataset with five different metadata (genres, tags, directors, actors and countries) shows that our proposed ensemble algorithms achieve a considerable 7% improvement in the Mean Average Precision even with state-of-art collaborative filtering algorithms.