Editorial Summary

A workflow for the automated generation of databases from literature 

Since the dawn of modern science, there has been a continuous, exponential growth in the volume of published scientific literature. Materials data can provide the foundation for models and theories to navigate the physical/chemical space and ultimately drive discovery. However, access to information from unstructured literature at such a massive scale presents significant technical and practical challenges. In general, curated databases are scarce and often limited to theoretical data only. Large-scale theoretical datasets have been proven to be a revolutionary tool in the search for new materials with unique properties and for the discovery of intricate materials trends. Furthermore, theory datasets have been a platform for constructing machine learning models with enhanced throughput. However, databases containing experimental results are much rarer and typically smaller. The existing landscape of experimental datasets is incomplete and fragmented, and most of the known experimental results remain accessible only through unstructured scientific literature. In this work, a team led by Prof. Stefano Sanvito from the School of Physics, Trinity College, Dublin, Ireland, presented a workflow based on the fine-tuning of BERT models for different downstream tasks, which results in the automated extraction of structured information from unstructured natural language in scientific literature. Contrary to existing methods for the automated extraction of structured compound-property relations from similar sources, this workflow does not rely on the definition of intricate grammar rules. Hence, it can be adapted to a new task without requiring extensive implementation efforts and knowledge. The authors tested their data-extraction workflow by automatically generating a database for Curie temperatures and one for band gaps. These were then compared with manually curated datasets and with those obtained with a state-of-the-art rule-based method. Furthermore, in order to showcase the practical utility of the automatically extracted data in a material-design workflow, the authors employed them to construct machine-learning models to predict Curie temperatures and band gaps. In general, although more noisy, automatically extracted datasets can grow fast in volume and such volume partially compensates for the inaccuracy in downstream tasks.

一个从文献中自动生成数据库的工作流程