Exploring Trend Life Cycles in Science and Innovation through Text Mining

Abstract

Research topics, and the words used to describe them, rise and fall in popularity over time. The fastest rising topics are typically called trends or bursts—recent examples in computer science include deep learning, edge computing, and the internet of things. While individual researchers typically experience trends as an increasing series of mentions by colleagues, papers, and funding opportunities, they can also be empirically measured by looking at the frequency of terms in publications over time. When historical trends are measured this way and plotted on a graph, they appear to obey a common life cycle. However, there is no scholarly agreement on how this life cycle should be modelled. Previous work compared the performance of models using a handful of trends already known to the researchers. The small sample size and potential for selection bias makes it difficult to draw any firm conclusions. In this thesis, we combine automatic trend detection with mathematical modelling of trend life cycles to investigate the dynamics of trends in science and innovation at scale. Our main contributions are (a) a semi-automated pipeline to detect trends in large datasets of publications, (b) a comparison between two popular life cycle models on a dataset of automatically selected trends across the fields of computer science, particle physics, mental health, and cancer research, (c) a demonstration that a random forest classifier can predict whether detected trends in research will rise and fall in the future, and (d) an investigation of lead-lag relationships between trends in papers, patents, and grants. This thesis advances understanding of how big ideas in research emerge, grow, and decline, and the temporal relationship between trends in research and innovation.

Date of Award	1 Aug 2022
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Goran Nenadic (Supervisor) & Robert Stevens (Supervisor)