Recently we've started a new project on context aware-multimedia recommendation. The projects is still in the startup phase, nevertheless we're collecting and going through relevant litterature.

There are huge amounts of research being conducted in the field of recommender systems, the amount of research on the topic of context-aware recommender systems is considerably smaller. One of the very few events that have been held on this topic was last year's Workshop on Context-aware Recommender Systems (CARS-2009) at the ACM Recommender Systems 2009 (RecSys09) conference. I had the opportunity to attend the conference, but was sadly unable to attend the workshop. The papers presented at the workshop, all available on the website, all stress the importance of context-awareness.

All recommender systems are context-dependent, why nobody seems to be aware of that (or at least choses to ignore it) is a mystery to me. What does the rest of the RecSys community do, what are the motivation behind not using context? As the papers from CARS show, context-awareness does not have to be complicated, simple contextual assumptions can increase recommendation results considerably, still, looking at recsys tracks and papers at other conferences, you're more likely not to find anything on context-awareness than the other way around.

I do believe this is not due to ignorance or disinterest from the community. The reason, I think, is that this concept has been overlooked, we're all blinded by all the different collaborative filtering approach and their extensions. Often, when there is a paper on context-awareness, the authors will not even have considered their work to be related to the context-aware topic, instead they use terms such as hybrid, higher-order, temporal, etc. Making a simple search for any of those terms on either the ACM portal, Google Scholar or any related library you'll find hundreds of papers, none of them actually stating the simple fact that they deal with context-awareness.


Leave a Comment