Towards Expert Preference on Academic Article Recommendation Using Bibliometric Networks

Expert knowledge can be valuable for academic article recommendation, however, hiring domain experts for this purpose is rather expensive as it is extremely demanding for human to deal with a large volume of academic publications. Therefore, developing an article ranking method which can automatically provide recommendations that are close to expert decisions is needed. Many algorithms have been proposed to rank articles but pursuing quality article recommendations that approximate to expert decisions has hardly been considered. In this study, domain expert decisions on recommending quality articles are investigated. Specifically, we hire domain experts to mark articles and a comprehensive correlation analysis is then performed between the ranking results generated by the experts and state-of-the-art automatic ranking algorithms. In addition, we propose a computational model using heterogeneous bibliometric networks to approximate human expert decisions. The model takes into account paper citations, semantic and network-level similarities amongst papers, authorship, venues, publishing time, and the relationships amongst them to approximate human decision-making factors. Results demonstrate that the proposed model is able to effectively achieve human expert-alike decisions on recommending quality articles.