Yesterday I attended a workshop that was aimed at fostering research collaboration between our department and BSkyB. After a short introduction by the head of the department, a number of members of staff gave short (10 minute) pitches about their past and current research, and areas they are interested in for potential collaboration. The range of work being done in the department is huge- perhaps this deserves a post of its own.

I presented (along with one of my supervisors Licia Capra) on future recommender systems. Given that I had less than 10 minutes to talk on this incredibly broad and interesting subject, there was no way I could say everything I wanted. So I tried to compress my ideas into four key points:

1. Motivating Recommender Systems. This part sounds obvious; I’ve read countless papers that talk about information overload, a term that comes from a book (with other theories of arguable value?). However, as the interest in recommender systems grows and the research fields matures, the motivation itself seems to be shifting: we are leaving behind (traditional) approaches to search and web interaction and heading toward an age of information personalisation and discovery. I played on this idea by modifying the google home page (see slide 2 of my presentation).
2. Born on the Web. When we think of recommender systems, we think of the web. We think about the wisdom of the crowds. The typical example thrown around in the pub (to non-researchers) is Amazon. To highlight the flood of online systems, I borrowed a slide from Oscar Celma’s music recommendation tutorial. So: what does this mean? Interacting with a recommender system is a lonely experience; we sit alone in front of a computer screen. We receive recommendations for ourselves, based on our profile. I echoed the “users who have been like-minded in the past will be like-minded in the future…”; assumption that we build these systems on. The key is that the data that traditional systems have been handling assumes individual profiles, an early warning for my next point:
3. Future Systems. Recommender systems centre on music, movies, and (in general) e-commerce items. So which way are things moving? Already there is exciting potential for recommending people, which moves away from the traditional way things work. Search engine prototypes are making it a social experience. What else? I split this part into two halves:

• Leaving the Web Behind. The age of (information) discovery need not reside only on the web. What about recommending television (like the RecSys Strands Startup Winners, and work being done at Telefonica I+D)? What about discovering content when on the move (like work being done by these two guys)? In other words, heads up for ubiquitous recommender systems. We are going to be building systems that mirror the interaction that has been happening online. Next thing you know your fridge will be recommending what the best dinner you can make is.
• Emerging from the Web. On the other hand, a few minutes browsing around show that, rather than building a world to mirror web interactions, the systems online may move outward. Google plans to take over tv, Telefonica’s Imagenio is growing, and p2p will kill traditional television (note the bias to tv: remember who I was presenting to!)

What does this mean? Well, it means a lot of things. But the point of most interest (to me) is that both the context of recommendation and context for recommendation are changing. Context Of? We will be in different places, looking for different things, with different people and friends (see a recent tutorial at RecSys by Amodavicius). Context for? We will be seeking recommendations at different times (and don’t tell me that having a hard disk on your television means you won’t want to see the world cup final live!), we will be seeking recommendations for groups (recommend to a family?). We need finer grained models of users, in environments where getting this kind of information is more difficult.

This is just a brain dump of what’s left after my rushed talk. Any thoughts?