Presentation Open Access

RISIS Research Seminar 16/ Link prediction in knowledge networks using exogenous and endogenous attributes: a machine learning approach

Giovanni Cerulli; Antonio Zinilli

The researchers propose a supervised machine learning approach to predict partnership formation between universities. The focus is  on successful joint R&D projects funded by Horizon 2020 programme in three research domains: Social Sciences and Humanities, Physical and Engineering Sciences, and Life Sciences.


The researchers perform two connected analyses: link formation prediction, and feature importance detection. As for link prediction, using out-of-sample cross-validated accuracy and a set of network endogenous and exogenous attributes, the researchers obtain 90% prediction accuracy when both types of attributes are used, and around 65% when using only the exogenous ones. This proves that partnership predictive power is on average 25% larger for universities already incumbent in the programme than for newcomers. As for feature importance, by computing super-learner average partial effects and elasticities, the study finds that the endogenous attributes are the most relevant in affecting the probability to generate a link and observe a largely negative elasticity of the link probability to feature changes, fairly uniform across attributes and domains.

Files (1.6 MB)
Name Size
1.6 MB Download
All versions This version
Views 175175
Downloads 3535
Data volume 55.0 MB55.0 MB
Unique views 165165
Unique downloads 3333


Cite as