Automatic Construction of Knowledge Graphs from Scientific Literature for Copper-based Composites

Providing knowledge graphs for materials science facilitates understanding of the key data such as materials structure, property, etc. and their relations. However, very little work has been devoted to it. Meanwhile, immediately applying machine learning to materials computation still suffers from the lack of data and costly acquiring. To tackle these problems, we propose literature-aid automatic entity and relation extraction by deliberatively designed matching rules, especially for copper-based composites. Next, we fuse the knowledge by calculating the semantics similarity. Finally, the materials knowledge graphs are constructed and visualized on the Neo4j graph database. The experimental results show a total of 6,154 entities and 15,561 pairs of relations are extracted on the 69,600 open-accessed documents of copper-based composites, with their precision and accuracy rates over 80%. Further, we exemplify the effectiveness by building materials structure-property-value meta-paths and analyzing their impacts.