Impactrank: A Study on News Impact Forecasting

In this paper we developed a framework and a measure for news impact forecasting. We proved the viability of our impact forecasting approach using a SVM based forecaster on six months of NYT corpus - consisting of 16,852 articles. We experimented with different feature selection and ranking algorithms including standard frequency based methods, as well as a new method named ImpactRank. Our ImpactRank based forecaster performed as the best feature ranking technique while providing a graph suitable for browsing and identifying the most influential topics, entities and inter-relationships going into its impact predictions.