Experimental Evaluation of Academic Collaboration Recommendation Using Factorial Design

Abstract

Recommender systems have been very used in e-commerce and online social networks. Among various challenges to construct such systems, how to parameterize them and their evaluations are two vaguely explored issues. Generally, each recommendation strategy has parameters and factors that can be varied. In this article, we propose to evaluate the impact of key parameters of two state-of-the-art functions that recommend academic collaborations. Our experimental results show that the factors affect recall, novelty, diversity and coverage of the recommendations in different ways. Finally, such evaluation shows the importance of studying the impact of the factors and factor interactions in the academic collaboration recommendations context.