Table-Based Text Parsing Methods for Polymer Science: A Comparative Study

Polymer informatics has emerged as an area of interest in the search for more sustainable plastics. Yet, there is a dearth in large, high-quality, polymer-based materials databases that are open-source for the polymer research community. Efforts to curate such databases involve autogenerating large experimental materials databases by mining chemical data from scientific literature, with a specific focus on mining tabular data as these are particularly rich sources of polymeric information. This study compares the performance of two tools in extracting large volumes of tabular data about polymer names and their glass-transition, melting and decomposition temperatures: a table-extraction tool that employs a downstream neural network to resolve polymer names from the table fields, and the table-mining part of the "chemistry-aware" natural-language-processing tool, ChemDataExtractor. We find that both methods afford high precision. The recall of the former is boosted by preserving some of the implicit structure of the source tables, while the latter offers a far wider scope of knowledge extraction from the literature on polymer science.