Disciplinary assessment of scientific content by higher-order citation mining.

ABSTRACT Discovery of disciplinary fields, their relationships, and key contributors is an important research field of modern scientometrics. In this contribution, we study the problem of disciplinary/interdisciplinary interpretation of scientific content in the context of higher-order citation graphs. In doing so, we introduce the general approach for representing scientific publishers (conferences, journals) and their relationships by means of multi-dimensional tensor models. Consequently, tensor decomposition methods allow us to identify scientific fields in terms of key authorities, hubs and topical context expressed by keywords and/or fields of study. The proposed strategy of tensor importance sampling allows for efficient and effective analysis of large-scale publication databases in state-of-the art cloud computing infrastructures. Consequently, we present results of our large-scale evaluations for the Microsoft Academic Graph dataset with various tensor modeling strategies, in direct comparison with state of the art content mining and graph mining methods common to the scientometric domain. The comprehensive comparative analysis shows that tensor-based interpretations show considerable agreement with established content mining methods and graph-based authority ranking. At the same time, tensor based analysis allows for discovery of novel inter-disciplinary and trans-disciplinary fields, which are focused on subjects of investigation rather than on established methodologies and strongly connected research communities. In this sense, our approach complements existing mining methods for scientific content, and allows for novel insights into research collaboration patterns. [ABSTRACT FROM AUTHOR]