Metadata in movies recommendation: a comparison among different approaches

• Lucas Tobal Percevali USP
• Marcelo Garcia Manzato USP

Abstract

This paper proposes a study and comparison among a variety of metadata types in order to identify the most relevant pieces of information in movie recommendation. We used three algorithms available in the literature to analyze the descriptions, and compared each other using the metadata extracted from two datasets, namely MovieLens and IMDB. As a result of our evaluation, we found out that the movies’ genres are the kind of description that generates better predictions for the considered content-based recommenders.

Keywords: recommender systems, metadata, matrix factorization, latent factors