The Use of Dissimilarity Measures for the Study of Evolution in Scientific Fields

The scientific theory of evolution by natural selection began with Charles Darwin’s On the Origin of Species published in 1859 [1,2]. Evolution as a scientific theory has been used in many other disciplines as well, including medicine [3,4], psychology [5,6], anthropology [6], forensics [7], agriculture [8], and other socialcultural applications [9,10]. In this paper, we aim at studying the evolution of scientific fields by investigating their developmental trends shown in scientific literature [11].

Evolution in scientific fields has been drawing attention among researchers and scientists in recent years. One of the main issues in evolution of scientific field is to quantify the dissimilarity between two collections of scientific publications in literature. The temporal evolution of scientific research can be observed in retrospective studies in many fields. What is more, the “evolution map” of scientific fields helps us understand the nature of scientific development and the relative importance of different topics or publications [11]. Innovations and scientific breakthroughs keep on emerging, leading to new or improved technology and scientific findings, which, then, shape new developmental trends in various research areas. One of the main issues in evolution of scientific field is to quantify the dissimilarity between two collections of scientific publications in literature. There are many dissimilarity measures encountered in many different areas such as biology, computer science, mathematics, psychology, statistics, etc. Finding the appropriate measures among such a collection of choices is of fundamental importance to pattern classification, clustering, and information retrieval problems [12-14].

Over the years, there have been many approaches for measuring scientific evolution. For instance, Vargas-Quesada et al. [15] introduced a graphic representation of the intellectual structures in the form of scientograms using scientometric information such as cocitation network from a certain scientific domain. Their approach allows one to detect patterns and tendencies of scientific evolution in a scientific domain through network visualizations. Dias et al. [16] used an information-theoretic measure of linguistic similarity to investigate the organization and evolution of scientific fields based on the 20,000 most frequent words from the abstracts of the papers considered excluding a list of stop words. Jurgens et al. [17] proposed a method on measuring scientific evolution by studying how scientific works frame their contributions through different types of citations and how this framing affects the field. Frank et al. [18] used the Microsoft Academic Graph to study the bibliometric evolution of AI research and its related fields from 1950 to 2019. The problem with these works is that they only adopted few (mostly just one single) dissimilarity measures in their study, ignoring the fact that there are many such measures [19,20].

In recent years, many attempts on integrating different dissimilarity measures have been made and they show promising results in various areas such as image classification [21], text categorization [22], and patten recognition [23]. Most of these works combine dissimilarity measures using a weighted sum with weights determined using different algorithms such as trialand- error method, and cross validation. In the study of scientific evolution, Zheng & Jiang [24] proposed a novel approach for the integration of twelve dissimilarity measures based on keywords distributions in scientific fields using principal component analysis (PCA). They collected a collection of bibliographic records of articles from four selected scientific fields published from 1991 to 2019 and obtain the yearly keyword distributions and calculated the values of twelve dissimilarity measures between the keyword distributions for each pair of successive years. Then PCA is unutilized to combine these dissimilarity measures. Their results show a decreasing trend for the evolution between two successive years in all chosen fields during the time 1991-2019 [25,26].

Most of the studies on scientific evolution have been limited to the use of single measures. Considering the successful applications of integrating dissimilarity measures, it would be a good idea to study the use of different integration techniques (e.g., ensemble methods) on scientific evolution. We believe that more efforts are needed to systematically study various properties of different dissimilarity measures in the study of scientific evolution. Future works are also needed to explore and compare the advantages and limitations of different integration approaches in this area.

None.

No conflict of interest.