Time dependency in TV viewer clustering

Web-based catch-up TV services allow users to watch programs at their favoured time and device and are revolutionizing the existing TV watching habits. With the increasing offer and demand for catch-up TV, it has become evident that there is a need for personalised recommendations that will help users to pick programs of interest from a large collection of available content. In order to mitigate the cold start problem, a catch-up TV recommender needs to exploit information pertaining to the watching patterns and stereotypical behaviour of users. This paper presents an exploratory study into the watching patterns and stability of the identified stereotypical user behavior using a
large-scale dataset gathered by an Australian catch-up TV services provider.
Using clustering, we were able to identify eight distinct and meaningful behaviour stereotypes. We further analysed these clusters and found that clusters with highly dominant watching patterns stabilise sooner and can be identified more accurately than others. Our work provides a solid foundation for developing future catch-up TV recommender systems.