A Framework for Predicting Impactful Research in Hydrological Science

Global dependence on citation-based measures of scientific research impact is perpetuating long-existing academic biases. Stakeholders of hydrological research (e.g., grant funders and program managers) often make crucial decisions about directing their resources towards certain research endeavors based on popularity, including citation records, of previous research. We propose a framework for assessing the future impact of hydrological research that takes into account multiple (additional) factors of impact including probabilistic distributions of subtopics within research papers, interdisciplinarity of topics in publication, venues of publication, direct social (research-into-use) implications, and author credibility. We use Natural Language Processing (NLP) to leverage large amounts of textual data in combination with quantitative data and an ensemble of models (Latent Dirichlet Allocation, Attention and Random Forest Regression) to generate democratic normalized impact scores. Preliminary results suggest our framework is able to capture early-warning signals for impactful hydrological research. Our goal with this project is to help hydrology become a more democratic science.