Tag-based Recommender System

Abstract

Organizers of STRP Art and Technology Festival want to enhance the festival experience of visitors whilst also learning more about these visitors. The proposed solution is a tag-based recommender system, where the feedback received from visitors will allow STRP to learn more about how visitors perceive art pieces and in turn provide visitors with recommendations of other art pieces to view, at the festival. During the course of this thesis, we first explore how we can learn about the preferences of visitors using the tags they contribute, paired with a rating for the art piece. We do this by investigating a semantic mapping tool, Relco from TU Eindhoven, which enables us to map tags to concepts from a vocabulary, with the help of a lexical ontology such as WordNet. We experiment with the stemming of tags before using them in string matching algorithms. Further, we investigate how influential the vocabulary is onto which we map tags, when using Relco. Finally, we evaluate recommendation algorithms. We explore collaborative, content-based and hybrid forms of recommendations. We conclude that, in this context, content-based recommendation algorithms perform the most accurately and consistently. We also conclude that the semantic extension in a tag-based recommendation algorithm enables us to accurately profile a visitor and art piece with a finite number of concepts.