Unlocking New Insights for Electrocatalyst Design: A Unique Data Science Workflow Leveraging Internet-Sourced Big Data

In the past few decades, numerous electrocatalyst design studies have been reported. Although machine learning (ML) has recently emerged as a more efficient alternative to traditional trial-and-error methods, the cost of preparing training data remains high. Inspired by the success of models like ChatGPT, which learns from a vast corpus of text data collected from the internet, we developed a data science workflow initiated by collecting datasets via a highly automated web crawler. We trained artificial neural network models with acceptable accuracy in predicting electrocatalytic performances and used black-box interpretation methods to mine universal material design knowledge, verifying model reliabilities with data collected from as many as 5277 publications. Thoughtfully, we introduced transfer learning (TL) to address the data scarcity issue for electrocatalysts in neutral electrolytes, with fewer available publications. TL could provide reliable optimization advice even in unknown areas, with knowledge transferred from similar fields. This study examined the patterns of numerous previous electrocatalysts from a data science perspective and proposed a universal ML paradigm to assist in the design of unique materials based on transferable big scientific data.