A Model of Research Collaboration Analysis Based on Publication Similarity

Abstract

This research proposes a model for analyzing what are the suitable parts of papers used to calculate the research similarity between a pair of researchers. First of all, a set of published papers based on six sub-fields of computer are selected from SCOPUS bibliographic database. Next, nouns in such papers are extracted and reduced to stem nouns. The words are submitted to Author Topic Model technique. The outputs of model are probability distributions over six topics for a particular researcher. For calculating the research knowledge similarity between two researchers, cosine similarity is applied to their probability distributions. The experiments show that these three factors influence to the right collaboration in test data: (1) recent year of publication; (2) shorter degree of separation; (3) multiple part of papers including abstract and reference sections.