Content area
Full text
1. Introduction
Scientific output has accelerated at an exponential pace owing to digital technology. This trend is particularly noticeable through the significant growth in digital scholarly publications across every discipline (Khabsa and Giles, 2014; Ponte et al., 2017; Taubert and Weingart, 2017). The increase in digitally accessible scholarship is exciting although this growth presents researchers with a daunting challenge as they seek to grapple with and benefit from the expanding corpus of information. This growing challenge is particularly prevalent in materials science (Mysore et al., 2017; Weston et al., 2019), where researchers seek to discover new recipes for improving materials performance.
To further explain the problem, digital scholarly publications, as a form of scholarly big data (Tuarob et al., 2016), generally describe the research design (method, sample and treatment), report finding and state conclusions. Specific to the case of materials science, results buried in a digital publication may help a researcher predict a material’s performance for future work. For example, published results may report that “a specific recipe of magnesium, copper, and yttrium (Mg-Cu-Y), followed by heating and cooling these combined alloys at a designated temperature, results in a certain grade of metallic glass.” The common approach for extracting this knowledge from digital text is extremely time-consuming. At a high level, a researcher needs to identify, locate and access relevant scholarly resources, such as peer reviewed journal publications, conferences papers or patent documents, from the ever-expanding body of digital scholarship. Next, the researcher needs to read and manually extract key knowledge. Weston et al. (2019) emphasized that it is impossible for a researcher to read and extract knowledge contained in the vast, expanding store of published research and referred to this problem as a bottleneck in materials discovery. This challenge underscores the need to explore machine driven approaches for extracting expert knowledge recorded in published research.
The research presented in this paper is motivated by both the need and the opportunity to advance knowledge extraction techniques for materials science scholarly resources. Interconnected to this goal is the need for researchers to understand levels of knowledge extraction, as well as the strengths, limitations and potential application of these varied approaches for materials science. The research presented here aims to also contribute...