Speed of Research Topic Evolution and Scientific Performance:Evidence from Computer Science
[Purpose/Significance]Using deep learning techniques,this study quantifies the speed of topics evolution for scientists over time and explores the historical development and heterogeneity of this trend,as well as the relationship between it and scientific performance.[Method/Process]First,based on publication data from nearly one million computer scientists worldwide from 1980 to 2019,it adopted Doc2vec to measure the distance between text features in sets of papers produced by scientists in adjacent years and calculated the speed of topic evolution.Next,it explored the relationship be-tween the topic evolution speed and scientific performance,and compared the differences in dynamic development of topic evolution among scientists in different career stages.[Result/Conclusion]Empirical evidence indicates that,over the past four decades,the speed of research topic evolution among scientists has gradually decreased,with the global trend in com-puter science to"exploitation-oriented research".Lower speed of topic evolution can bring the optimal research performance,and the relationship between them shows an inverted U-shaped curve.In the career lifecycle,elite scientists tend to have a gradually decreasing level of topic evolution,showing a trend of focusing on specific topics,while that of non-elite scientists increasing gradually or remaining steadily in most of the period.The empirical results provide policy guidance for assess-ment of discipline development,scientists'career advancement,and science policy formulation.
NETL
NSTL
万方数据
