Mining Semantic Relations in Data References to Understand the Roles of Research Data in Academic Literature

Research data serves important roles in scientific discovery and academic innovation. To appropriately assign credit for data work and to measure the value of research data, it is essential to articulate how data are actually used in research. We leveraged a combination of computational methods and human analysis to characterize different types of data use by mining semantic relations from the phrases where data are referenced in academic literature. In particular, we investigated references to data in the bibliography of a large social science data archive, the Inter-university Consortium for Political and Social Research (ICPSR). After retrieving and extracting semantic relations as subject-relation-object triples, we used rule-based methods to classify them. We then annotated samples from 11 frequent classes of data reference triples and found that they vary primarily along two dimensions of data use: proximity and function. Proximity describes the distance between the author and the data they reference (e.g., direct or indirect engagement). Function describes the role that data plays in each reference (e.g., describing interaction or providing context). These semantic relationships between authors and data reveal the ways data are used in scientific publications. Evidence of the variety of ways data are used can help stakeholders in research data curation and stewardship - including data providers, data curators, and data users - recognize the myriad ways that their investments in data sharing are realized.