I'm going into my third year of studies as an AI student and am planning my third year project. I have been considering a recommendation system of some sort. The motivation for this is to gain an understanding of how people evaluate products (what makes the products desirable) and consequently attempt to build a system that would understand this. Currently my thinking is along the lines of a system that would be able to differentiate between different priorities in peoples' likes and dislikes. For instance a person who is environmentally very aware probably wouldn't want to buy products that are not.
So the question is - What things are most in need of repair/development in the modern web AI systems (Google, Amazon, Last.fm and so on).
My project is limited to about 6 months but I would be interested to hear any thought on the subject.