Natural materials are able to achieve a wide range and combination of properties through the arrangement of the material’s components. These biological materials are often more effective and better suited to their function than engineered materials, even with the use of a limited set of components. By mimicking a biological material’s component arrangement, or structure, man-made bioinspired materials can achieve improved properties as well. While considerable research has been conducted on biological materials, identifying the beneficial structural design principles can be time-intensive for a materials designer. Previously, a text mining algorithm and tool were developed to quickly extract passages describing property-specific structural design principles from a corpus of materials journals. Although the tool identified over 90% of the principles (recall), many irrelevant passages were returned as well with approximately 32% of the passages being useful (precision). This paper discusses approaches to refine the program in order to improve precision. The text classification techniques of machine learning classifiers, statistical features, and part-of-speech analyses, are evaluated for effectiveness in sorting passages into relevant and irrelevant classes. Manual identification of patterns in the returned passages is also employed to create a rule-based method, resulting in an updated algorithm. An evaluation comparing the revised algorithm to the previously developed algorithm is completed using a new set of journal articles. Although the revised algorithm’s recall was reduced to 80%, the precision increased to 45% and the number of returned passages was reduced by 22%, allowing a materials designer to more quickly identify potentially useful structures. The paper concludes with suggestions to improve the program’s usefulness and scope for future work.

This content is only available via PDF.
You do not currently have access to this content.