Wandering between close and apart: are scholarly domains crossable?

Abstract

Interdisciplinary collaborations, i.e., scholarly crossdomain collaborations have generated huge impact to society, and has been previously proved to exhibit domain skewness[20]. To illustrate, scholarly cross-domain collaborations seldom emerge between irrelevant scholarly domains. In this work we address a question to determine the possible existence of scholarly cross-domain collaborations, namely: "Are scholarly domains really crossable?" Using a real-world scholarly dataset, i.e., Microsoft Academic Graph (MAG)[1] with 126 million papers collected from 53,834 domains, we take the initiative to formalize a "crossability" quantification problem, where the "crossability" serves as an index that aims to evaluate the ability of two scientific domains to establish collaborations. In doing so, we propose two metrics, i.e., co-paper ratio and hierarchical distance, where the former one is the ratio of common papers in two domains to which in a single domain, and the later one is the difference of domains' levels according to their positions in the hierarchical structure. Interestingly, we observe a peak pattern, meaning that the influence of research work, i.e., number of citations, climbs to a peak when its domains' count goes to a certain number, after which the citation count decrease sharply. Our discovery indicates that a moderate amount of domain "crossability" helps to improve the impact of research work, which, however, could be weakened under excessive "crossability". With elaborately modeling, we reproduce this peak pattern and briefly discuss the reason of the existence of peak.