IRIS: Learning the Underlying Information of Scientific Research Interests Using Heterogeneous Network Representation

Abstract

Understanding scientific research fields and finding potential relations between seemingly distinct fields can help researchers rapidly grasp their most interested topics with expertises. In this study, we construct a heterogeneous network which contains authors, keywords, papers and institutions, and built an “Integrated Research Interest Space (IRIS)” which can represent both author and keyword nodes. Similar keywords in the sense of research interest and research manner can obvious aggregate together. Authors that are interested in different keywords distributed in different IRIS areas, with strongly associated with research objectives and methodologies of the keywords. The average similarities between authors and their real used keywords is significantly higher than that of randomly chosen author-keyword pairs. Based on these observations, we propose a simple algorithm which attempts to recommend potential interested keywords for researchers, and got meaningful results. Our study may also give useful hints for understanding research interests and discovering potential cross disciplines.