If any collaborative filtering system for Usenet Net News is ever to be successful, it will need to meet the requirements of not only the users, but also the Net News system administrators who will have to build, install and maintain the system. These people are often volunteers or have been volunteered for the job, so our system must place a minimal load on them.
Net News systems consist of two primary parts: client news readers which are short lived programs started by individual users, and news server software which is constantly running. Many designs for a collaborative filtering system could be radically simplified if they had support from the news server software. Unfortunately, to be consistent with our goal of causing minimum hassle we have chosen to make changes only to the client news readers. There are three reasons for this.
First, informal opinions show that most news administrators have spent large quantities of time and chewing gum sticking together and customizing their news servers for their particular sites. The presence of these many code changes make the distribution of usable patches difficult and makes many administrators wary of making any further changes. News readers on the other hand are frequently compiled and built directly from distribution files with the only customizations being made in configuration files.
Second, news server processes are long lived and maintain large amounts of state. If the collaborative software were to cause a bug in a news server, large quantities of data might be irrecoverably lost. News readers on the other hand, are relatively state free. If the collaborative software caused a news reader to malfunction, it would be easy to back down and begin using an older version again.
Third, it was not our intention to carefully study the user interface issues of exactly how to present collaborative filtering information to users, or to how use the information to best filter and display articles. By making tools that help designers create news readers which support collaborative filtering, we aim to help others explore these human-computer interface issues.
We believe that placing the changes needed to support collaborative filtering in the client news reader programs will both minimize the hassle placed on system administrators and provide a more open system.