Year: 2012
Type: report
Abstract: Music recommenders often rely on experts to classify song facets like genre and mood, but user-generated folksonomies hold some advantages over expert classifications—folksonomies can reflect the same real-world vocabularies and categorizations that end users employ. We present an approach for using crowd-sourced common sense knowledge to structure user-generated music tags into a folksonomy, and describe how to use this approach to make music recommendations. We then empirically evaluate our “people-powered” structured content recommender against a more traditional recommender. Our results show that participants slightly preferred the unstructured recommender, rating more of its recommendations as “perfect” than they did for our approach. An exploration of the reasons behind participants’ ratings revealed that users behaved differently when tagging songs than when evaluating recommendations, and we discuss the implications of our results for future tagging and recommendation approaches.
Authors: J. Schleith, Simone Stumpf, Todd Kulesza
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2097198546