The diversity of canonical and ubiquitous progress in computer vision: A dynamic topic modeling approach

Abstract

Research trends are the keys for researchers to decide their research agenda. However, only a few works have tried to quantify how scholars follow the research trends. We address this question by proposing a novel measurement for quantifying how a scientific entity (paper or researcher) follows the hot topics in a research field. Based on extended dynamic topic modeling, the degree of hotness tracing of papers and scholars is explored from three perspectives: mainstream, short-term direction, and long-term direction. By analyzing a large-scale dataset containing more than 270,000 papers and 45,000 authors in Computer Vision (CV), we found that the authors’ orientation is more in the established mainstream rather than based on incremental directions and makes little difference in the choice of long-term or short-term direction. Moreover, we identified six groups of researchers in the CV community by clustering research behavior, who differed significantly in their patterns of orientation, topic selection, and impact. This study provides a new quantitative method for analyzing topic trends and scholars’ research interests, capturing the diversity of research behavior patterns to address the phenomenon of canonical and ubiquitous progress in research fields.