Improving Collaborative Personalized TV Services A Study of Implicit and Explicit User Profiling

摘要

As part of an ongoing research programme in personalized TV services, we have been developing a range of personalization and recommendation techniques that are well suited to the TV domain. In this paper we describe recent work on the application of data mining methods as a means of extracting programme similarity knowledge from shallow user profiles in order to address the sparsity problem normally associated with collaborative-filtering techniques. In particular, we compare the use of implicit behavioural profile data from Fischlar―a deployed online personal video recorder system―with the use of explicit user-rating profile data from PTVplus―a deployed online personal electronic programme guide―as employed in our earlier work. We evaluate the approach to show that it benefits from superior personalization accuracy across a range of experimental conditions, and that the benefits extend to implicit as well as explicit user profiling.