Electrocatalyst discovery through text mining and multi-objective optimization

TL;DR

This paper introduces a novel method combining text mining, word embeddings, and Pareto optimization to predict high-performance electrocatalysts, effectively utilizing scarce data sources and matching experimental results.

Contribution

It presents a new approach that leverages scientific texts and multi-objective optimization for electrocatalyst discovery, addressing data scarcity issues.

Findings

Text-based predictions align well with experimental electrochemical activity.

The method effectively narrows down candidate materials for specific reactions.

Text mining can complement traditional data sources in materials discovery.

Abstract

The discovery and optimization of high-performance materials is the basis for advancing energy conversion technologies. To understand composition-property relationships, all available data sources should be leveraged: experimental results, predictions from simulations, and latent knowledge from scientific texts. Among these three, text-based data sources are still not used to their full potential. We present an approach combining text mining, Word2Vec representations of materials and properties, and Pareto front analysis for the prediction of high-performance candidate materials for electrocatalysis in regions where other data sources are scarce or non-existent. Candidate compositions are evaluated on the basis of their similarity to the terms `conductivity' and `dielectric', which enables reaction-specific candidate composition predictions for oxygen reduction (ORR), hydrogen evolution (HER), and oxygen evolution (OER) reactions. This, combined with Pareto optimization, allows us to significantly reduce the pool of candidate compositions to high-performing compositions. Our predictions, which are purely based on text data, match the measured electrochemical activity very well.