Publication
Title
Recommending research collaborations using link prediction and random forest classifiers
Author
Abstract
We introduce a method to predict or recommend high-potential future (i.e., not yet realized) collaborations. The proposed method is based on a combination of link prediction and machine learning techniques. First, a weighted co-authorship network is constructed. We calculate scores for each node pair according to different measures called predictors. The resulting scores can be interpreted as indicative of the likelihood of future linkage for the given node pair. To determine the relative merit of each predictor, we train a random forest classifier on older data. The same classifier can then generate predictions for newer data. The top predictions are treated as recommendations for future collaboration. We apply the technique to research collaborations between cities in Africa, the Middle East and South-Asia, focusing on the topics of malaria and tuberculosis. Results show that the method yields accurate recommendations. Moreover, the method can be used to determine the relative strengths of each predictor.
Language
English
Source (journal)
Scientometrics: an international journal for all quantitative aspects of the science of science and science policy. - Amsterdam
Scientometrics: an international journal for all quantitative aspects of the science of science and science policy. - Amsterdam
Publication
Amsterdam : 2014
ISSN
0138-9130
Volume/pages
101:2(2014), p. 1461-1473
ISI
000343609900034
Full text (Publisher's DOI)
Full text (open access)
Full text (publisher's version - intranet only)
UAntwerpen
Faculty/Department
Research group
Publication type
Subject
Affiliation
Publications with a UAntwerp address
External links
Web of Science
Record
Identification
Creation 27.11.2014
Last edited 09.12.2017
To cite this reference