Identifying the “Ghost City” of domain topics in a keyword semantic space combining citations

Abstract

As an increasing number of scientific literature dataset are open access, more attention has gravitated to keyword analysis in many scientific fields. Traditional keyword analyses include the frequency based and the network based methods, both providing efficient mining techniques for identifying the representative keywords. The semantic meanings behind the keywords are important for understanding the research content. However, traditional keyword analysis methods pay scant attention to semantic meanings; the network based or frequency based methods as traditionally used, present limited semantic associations among the keywords. Moreover, the ways in which the semantic meanings behind the keywords are associated to the citations are not clear. Thus, we use the Google Word2Vec model to build word vectors and reduce them to a two-dimensional plane in a Voronoi diagram using the t-SNE algorithm, to link meanings with citations. The distance between semantic meanings of keywords in two-dimensional plane are similar to distances in geographical space, thus we introduce a geographic metaphor, “Ghost City” to describe the relationship between semantics and citations for hot topics that have recently become not so hot. Along with “Ghost City” zones, “Always Hot”, “Newly Emerging Hot”, and “Always Silent” areas are classified and mapped, describing the spatial heterogeneity and homogeneity of the semantic distribution of keywords cited in a domain database. Using a collection of “geographical natural hazard” literature datasets, we demonstrate that the proposed method and classification scheme can efficiently provide a unique viewpoint for interpreting the interaction between semantics and the citations, as “Ghost City”, “Always Hot”, “Newly Emerging Hot”, and “Always Silent” areas.