Web caching attempts to keep copies of popular content in a local cache. This should save the cost of a later download by the same or another user. Studies in world wide web (WWW) request patterns have indicated that 70% of requests from a web caching group are one time references. This implies that in a group of people there is very little overlap in the content they view. This thesis describes a system that helps increase the overlap of the pages viewed by a local cache group. It works by passive learning what each user is interested in and builds a small database of the sites that they find to be good references. It then shares these databases with other users with similar interests. This shows promise of becoming a much faster and more accurate watt of finding references compared to search engines. It also has application in e-commerce as on-line businesses are very interested in their visitors interests. (Abstract shortened by UMI.)