Predicting citation patterns: defining and determining influence

Definitions for influence in bibliometrics are surveyed and expanded upon in this work. On data composed of the union of DBLP and CiteSeer x , approximately 6 million publications, a relatively small number of features are developed to describe the set, including loyalty and community longevity , two novel features. These features are successfully used to predict the influential set of papers in a series of machine learning experiments. The most predictive features are highlighted and discussed.