Using Machine Learning to Uncover Latent Research Topics in Fishery Models

Abstract

Modeling has become the most commonly used method in fisheries science, with numerous types of models and approaches available today. The large variety of models and the overwhelming amount of scientific literature published yearly can make it difficult to effectively access and use the output of fisheries modeling publications. In particular, the underlying topic of an article cannot always be detected using keyword searches. As a consequence, identifying the developments and trends within fisheries modeling research can be challenging and time-consuming. This paper utilizes a machine learning algorithm to uncover hidden topics and subtopics from peer-reviewed fisheries modeling publications and identifies temporal trends using 22,236 full-text articles extracted from 13 top-tier fisheries journals from 1990 to 2016. Two modeling topics were discovered: estimation models (a topic that contains the idea of catch, effort, and abundance estimation) and stock assessment models (a topic on the assessment of the current state of a fishery and future projections of fish stock responses and management effects). The underlying modeling subtopics show a change in the research focus of modeling publications over the last 26 years.