Just another WordPress site - Ruhr-Universität Bochum

Abstract

Accelerating catalyst discovery is critical to advancing sustainable energy solutions, yet traditional experimental methods can be slow and costly. This study introduces a novel approach leveraging natural language processing (NLP) to expedite the discovery of high-performance catalysts by transforming scientific literature into actionable insights. Using a pipeline that integrates automated text mining with Word2Vec modeling, material properties are embedded as word vectors, allowing the extraction of latent knowledge from existing research. Three predictive models are presented, each trained on data from ternary alloy systems and applied to predict performance in a quaternary system, achieving significant improvements in predictive accuracy with NLP-derived material representations. Pareto front analysis further enhances the model’s ability to identify optimal catalysts, validated through experimental testing. This methodology showcases how NLP can serve as a powerful tool to bridge data gaps, guiding targeted experiments that reduce workload and accelerate breakthroughs in catalyst discovery. By transforming words into predictive knowledge, this approach presents a pathway to a more efficient and data-driven future in materials science.