Natural Language Processing for Earth resource management: a case of H2 Golden Retriever research

Abstract

Hydrogen is recognized as a crucial element in the future of green energy, playing a pivotal role in decarbonizing the economy. The pursuit of low-cost, high-performance, and durable materials for electrolysis has underscored the need for advanced tools to facilitate evidence-based decision-making in hydrogen research funding. This study introduces the H2 Golden Retriever (H2GR) system, a comprehensive platform leveraging Natural Language Processing (NLP), Knowledge Graph (KG), and Decision Intelligence for effective hydrogen knowledge discovery and representation. Hydrogen-related papers were systematically gathered from the web and subjected to preprocessing techniques such as noise and stop-word removal, language and spell checks, stemming, and lemmatization. NLP tasks included Named Entity Recognition using Stanford and Spacy NER, as well as topic modeling through Latent Dirichlet Allocation and term frequency-inverse document frequency (TF-IDF) analysis. The KG module facilitated the identification of meaningful entities, relationships, trends, and patterns within the realm of hydrogen research. The Decision Intelligence component created a simulation environment, capturing cost and quantity dependencies. The PageRank algorithm was employed to rank papers based on relevance. The H2GR system underwent random searches, yielding results that comprised a ranked list of papers, relevant entities, relationship graphs, an ontology of H2 production, and Causal Decision Diagrams illustrating component interactivity. Qualitative assessments by experts confirmed the satisfactory functionality of H2GR. This study demonstrates the significant potential of combining NLP and human-in-the-loop AI for accelerated knowledge discovery in hydrogen research. The adoption of an ontology centered on hydrogen production enables the identification of papers not highlighted by traditional citation-based metrics. H2GR presents a promising solution to alleviate the workload of experts in navigating the vast landscape of daily released hydrogen-related literature.