Modelling trend life cycles in scientific research using the Logistic and Gompertz equations

Scientific topics vary in popularity over time. In this paper, we model the life cycles of 200 trending topics by fitting the Logistic and Gompertz models to their frequency over time in published abstracts. Unlike other work, the topics we use are algorithmically extracted from large datasets of abstracts covering computer science, particle physics, cancer research, and mental health. We find that the Gompertz model produces lower median error, leading us to conclude that it is the more appropriate model. Since the Gompertz model is asymmetric, with a steep rise followed a long tail, this implies that scientific topics follow a similar trajectory. We also explore the case of double-peaking curves and find that in some cases, topics will peak multiple times as interest resurges. Finally, when looking at the different scientific disciplines, we find that the lifespan of topics is longer in some disciplines (e.g. cancer research and mental health) than it is others, which may indicate differences in research process and culture between these disciplines.