A Scocial Network based TV recommendation system

15.12.2005

Abstract
With the increasing amount of available TV-programs, finding what programs to watch and when to watch them can be quite problematic for the average television viewer. More and more people are starting to use Personal Video Recorders (PVR) to digitally record programs. I will in this document show the design for a plug-in for the Linux based PVR MythTV that will produce recommendations for TV-programs. I will also show that the peer-to-peer search infrastructure The Sozilized.net, a topology that tries to copy the way humans establish and maintain networks, is a good network platform for such a system and will also remove the use of a central server that most of today’s recommendations systems rely on.

I have by looking into analysis of recommendation systems found that item-based collaborative filtering provides the best recommendations when the data available is not too sparse. By using both collaborative filtering and content-based filtering depending on the data available the best recommendations will be produced. I will also show that using a semantic peer-to-peer system will contribute to making the recommendations process more efficient.

The number of people using PVRs has increased much in the later years and is very likely to do so in the coming years too. The available programs and channels are also increasing and I believe that recommendation systems for television are something that will be more and more common in the future.