A material dictionary database to extract information on permanent magnets from scientific articles

In this study, we developed a new information extraction method using a material dictionary database (MDDB), which parses scientific articles and collects related phrases from various ex-pressions. We used magnetic properties as an illustrative case to analyze the working of the proposed system. Structured terms comprising sub-phrases, tagged words, and their relationships enabled automatic annotation and information extraction. The MDDB was constructed on a pre-built knowledge base that includes information categories and related keywords. These cat-egories can be hierarchically structured and flexibly updated to extract a wide range of information on the associations between magnetic materials and properties along with the measurement systems used, structural analyses performed, and theoretical foundations. Herein, we propose preliminary rule-based phrase collection methods and label pattern extraction for phrases that can easily add new structured terms. We found 1,136 new phrases by label pattern-matching that enabled more related expressions to be retrieved from the text and enhanced the information extraction’s accuracy. Approximately 350 relationships among the material types, properties, and values were extracted from the manually modified annotations of 40 articles on permanent magnets. Our method can be applied to other research domains and can be used by such disciplines to build knowledge bases for any topic in their field.