Vineyard: a collaborative filtering service platform in distributed environment

As the amount of data shared on the Internet drastically increases, it becomes more important to utilize human feedbacks in retrieving information. While collaborative filtering of information is a promising approach to retrieve useful information, unfortunately there is an obstacle in the current collaborative filtering. Since a single system that serves for every type of object with optimal performance is unlikely, or at least likely to be too complicated as to implementation, it is more practical to use multiple systems each of which is dealing with a specific target. In many collaborative filtering systems, a system for book only deals with books, and another system for movie only deals with movies. The data stored in the former system is not accessible from the latter, and vice versa. This situation brings about inflexibility of collaborative filtering. We claim that this issue can be addressed by realizing multiple collaborative filtering systems on top of a single platform. Since all the profiles are shared on a single platform, they are accessible from all the systems. The filtering accuracy of our approach is identical with a certain kind of collaborative filtering system under a certain condition. We describe a design of collaborative filtering service platform in this paper. Our design of platform is fairly generalized, and it can be realized both in a centralized and peer-to-peer fashion.