HydroMind - An Interactive AI-aided Tool for Hydrology Literature

Abstract:

The number of articles published in Hydrology journals has increased exponentially in recent decades. Researchers, practitioners and stakeholders of hydrological science face an increasingly daunting task of synthesizing these publications efficiently, oftentimes becoming confined to only a slice of the broader state of science.

We tackle this issue with HydroMind, an interactive web-based tool for topic, time, and journal-based exploration of peer-reviewed publications in major hydrology journals (so far, we include Water Resources Research, Hydrology and Earth System Sciences, Journal of Hydrology, Hydrological Processes, Hydrological Sciences Journal, and Journal of Hydrometeorology from years 1991 to 2019).

Bibliometric analyses on 40,275 article-abstracts forms the foundation of this tool. We performed topic modeling with Latent Dirichlet Allocation (LDA) on these abstracts, identified topics through a hybrid objective-subjective approach, and used the posterior document-topic distributions to calculate the Jensen-Shannon distances between documents. The web interface allows the user to define and generate interactive, force-directed graphs based on topics, time and journal. Each node in the graph is a journal article, major topical themes (discovered via machine learning and natural language processing) are represented as colors and connections between articles, and distances represent topical similarity between papers and/or authors.

A diverse range of other analyses, such as temporal topic popularity trends, inter-topic correlations, journal diversities, which serve as ancillary information are derived from this analysis, and are available through the web-based tool. HydroMind is designed in a modular manner to incorporate more data, user options, and allows for modifications based on user feedback.

In the long term, we envision this tool as a part of a broad range of natural language processing-based tools which allow for fast, efficient contextual information retrieval from hydrology related textual data. These tools can also potentially benefit funding agencies, policymakers and first-order beneficiaries of hydrologic sciences in making relevant informed decisions.