Predicting News Story Importance Using Language Features

In this age of awareness, people have access to information like never before. Hundreds of newspapers and millions of bloggers present news and their interpretations in an openly accessible manner. With globalization, distant events can have impact on people thousands of miles away. While expert humans can recognize a potentially important piece of news, this is still a difficult problem for an automatic system. Since people are increasingly relying on multiple online sources of information, it is important to support users in filtering news automatically. In this work, we consider the problem of anticipating news story importance, i.e. given a news item, predicting if it will be of interest for a majority of users. Such ranking is currently done manually for newspapers, and we explore automatic approaches and indicative features for the same. Our main conclusion is that importance prediction is a hard problem, and pure textual features are not sufficient for classifiers with 90% accuracy.