A Hybrid Approach for Artwork Recommendation

In this article, we give an overview of the particularities that surround artwork Recommender Systems in museums. Unlike other domains, such as movies, where the user selects only one movie to watch, in the museum domain it is necessary to recommend a sequence of items to consume, in which the order and relationships among them might affect the recommendation. Furthermore, curators determine a place for each artwork in the museum, generating a distribution over the museum's rooms that imposes constraints about the way visitors walk through it. However, nowadays it is also possible to access to their digital collection, gaining independence from the physical distribution, allowing any visitor to be his/her own curator. Because of that, in this article we analyze and discuss different approaches to exploit the particularities of this domain. Specifically, we focus on one aspects, namely: item modeling. First, we summarize different approaches to represent and extract features from artworks that could be relevant for the recommendation task. Finally, we propose a novel approach to model artworks that could be useful in the recommendation task.