Snowball 2.0: Generic Material Data Parser for ChemDataExtractor

Area of Science:

• Materials Science
• Computational Chemistry
• Natural Language Processing

Background:

• The increasing volume of chemical data in scientific literature necessitates automated extraction methods.
• Software toolkits like ChemDataExtractor facilitate data extraction, creating a need for efficient text mining parsers.

Purpose of the Study:

• To introduce Snowball 2.0, a semisupervised machine learning sentence parser.
• To enhance automated chemical information extraction for data-driven materials discovery.

Main Methods:

• Developed Snowball 2.0, a semisupervised machine learning algorithm for sentence parsing.
• Validated the parser's performance using semiconductor band gap data from journal articles.
• Compared Snowball 2.0 against previous versions and integrated it into ChemDataExtractor 2.0.

Main Results:

• Snowball 2.0 demonstrates a 15-20% increase in recall compared to previous versions.
• Achieved improved performance in most configurations when integrated into ChemDataExtractor 2.0.
• Showcased better generalizability, learning efficiencies, and user-friendliness.

Conclusions:

• Snowball 2.0 offers advanced parsing capabilities for ChemDataExtractor, improving automated data extraction pipelines.
• The parser enables efficient extraction of chemical properties without additional training.
• Snowball 2.0 represents a significant advancement in text mining for materials science research.