Project: Pull Recommendations


Progress Report:
Abstract and references are above. I've also prepared and submitted my IRB forms, which includes the online survey that participants will complete.

Team: Megan Monroe

Based on my work with Jam My Jam, and my obsession with music recommendation, I would like to do a review/evaluation of some of the major music recommendation systems. My goal is to paint these systems as "push" recommendation systems. That is, systems in which current users receive new music ideas algorithmically or as direct messages from other users.

I'd like to argue then, that evaluating musical tastes, similar to image tagging and audio transcriptions, is a task that humans are inherently better at. A human, for example, could look at a window of the last 10 songs that a person listened to, and learn quite a bit about the listener, and how that listener's tastes relates to their own.

Finally, I'd like to propose strategies for harnessing this power in what would serve as a "pull" recommendation system. In this type of system, users would actively find new music themselves. The system would simply facilitate the search process.

Comment (Ben): This looks fine. I have a few questions though. Which systems will you evaluate and how will you evaluate them? Presumably the analysis will include both an interface and social analysis as well as actual content quality analysis. How will you objectively measure the quality of the recommendations?

As a scientific endeavor, I suggest that you position your thinking on this that you don't yet know whether humans are better or not. Or, perhaps there are ways in which humans are better and ways in which computers are better (i.e., I bet humans are not better are recommending things that they have never heard of). I'd like you to describe what the measures of success are, and then include humans in your assessment. Maybe you will be right, or maybe you won't.

Then, turning this all into a design proposal is good. Depending on the amount of effort that goes into the assessment part, I might expect that you go beyond the design phase, and also build some kind of prototype.

11/23/11: Comment (Ben)
I'm looking forward to seeing what you actually do here, and while I'm glad that you say you are making progress, there isn't much evidence here of it, so I remain guardedly optimistic.