Readers of big news sites might find useful the services of a
recommendation system, which, based on their previous choices, can
make recommendations from the latest articles. For testing and
training such recommendation systems, user-history data sets are
needed, that contain the reading history of users, and reflects their
preferences as closely as possible. But big news sites have many
articles, some emphasized heavily, and that biases the users choices,
and thus the preference-fidelity of the user-data. Browser-run
applications however can alter the appearance of the news sites, and
it can facilitate the easy finding of the articles that are really
interesting to the user.
This assignment's goal is to develop a browser-run application that,
firstly is capable of collecting user data and passing it on to a
server, and secondarily, it overwrite the typographic emphasis on the
news sites, and alters the arrangement of the articles in a way that
hopefully aids users in finding the articles that are interesting to
them. Data collected with this tool will hopefully give a more close
indication of the user tastes, and therefore recommendation systems
training on that data will give better recommendations.
In this paper I am going to present the steps of designing and
implementing such an application, and also study the difficulties met
and their solutions.