Friday, October 19, 2007

Seminar at TsingHua University

I did not go to Beijing for leisure. Actually I went there to present a paper at an international conference, and to pay visits to Dr. Zhang Li and Dr. Shi Yuanchun at TsingHua University. The pictures posted here were taken after work.

While at Tsinghua University, I gave a seminar on our own work on Recommender Systems to a group of research students. Recommender systems are used in places like Amazon. When you examine a book at Amazon, it informs you that “customers who bought this item also bought …” That’s an example of a recommender system. Some of them are based on the idea that people who are interested in the same items share some common interest, and you are likely to be interested in the other items that they were interested in. That’s call collaborative filtering. We are doing some work that can make use of the sentiments expressed in textual reviews to make recommendations, and that's what I presented at Tsinghua.


Liu Yunxia said...

We happened to discuss this kind of things in class yesterday. One problem about this is that people will buy something for others as a gift, but the system thinks it is the buyer who was interested in this and keeps recommending similar items which the buyer isn't really like himself.

StephenC said...

That's a good point, and a recognized problem in recommender systems. To some extent, it is still a kind of interest of the buyer, although it is indirect. To some extent it can be accounted for by a degree of interest, estimated by the frequency of purchases, etc.

A related problem is the difficulty in establishing the identity of the person expressing an interest or opinion. For example, often only the IP address of the machine is known.