The most visible parts of the system we implemented are the news reader clients used by users to read their news. So long as it is working properly, the remainder of the collaborative filtering system sits out of sight of the end users. We had originally chosen the xrn and nn news readers to gather usage statistics from because they are both popular news readers. Together, they also cover a range of interfaces from the text-based interface used by nn to the graphical user interface of the X based xrn. For these same reasons as well as our new found familiarity with their source code, we chose to modify nn and xrn to support collaborative filtering.
In modifying the news reader clients, our primary goal was to make it as easy as possible for users to vote and filter. As we previously explained, users spend so little time looking at individual articles, that we wanted to streamline the voting process into their natural news reading pattern. At the same time however, we wanted to subtly remind users that they could, and should, vote for or against the articles that they liked and didn't like. Finding efficient ways to get votes from users and present filtering information to them was the key challenge of modifying the news readers.