A supervised machine learning classification algorithm for research articles

Abstract

The issue of the automatic classification of research articles into one or more fields of science is of primary importance for scientific databases and digital libraries. A sophisticated classification strategy renders searching more effective and assists the users in locating similar relevant items. Although the most publishing services require from the authors to categorize their articles themselves, there are still cases where older documents remain unclassified, or the taxonomy changes over time. In this work we attempt to address this interesting problem by introducing a machine learning algorithm which combines several parameters and meta-data of a research article. In particular, our model exploits the training set to correlate keywords, authors, co-authorship, and publishing journals to a number of labels of the taxonomy. In the sequel, it applies this information to classify the rest of the documents. The experiments we have conducted with a large dataset comprised of about 1,5 million articles, demonstrate that in this specific application, our model outperforms the AdaBoost.MH and SVM methods. Copyright 2013 ACM.