Evaluation of Tag Clusterings for User Profiling in Movie Recommendation

Guglielmo Faggioli, Mirko Polato, Ivano Lauriola, Fabio Aiolli

Abstract: In the web 2.0 era, tags provide an effective mechanism to rapidly annotate and categorize items. However, tags suffer from many problems typically linked to language, like synonymy, polysemy, and ambiguity in general. To overcome this limitation, tag clustering can be used to group tags that represent similar concepts. One of the domains where tag clustering has shown to be particularly useful is the movie recommendation, where tags are used to represent users’ preferences and affinities. In this context it is not yet available a golden standard that can prove the quality of a clustering technique, especially considering that the final aim is the users’ satisfaction rather than an accuracy-like score. To this end, we propose an evaluation criterion for the quality of the resulting clusters based on human judgments.