Detecting Informal Data References in Academic Literature

Abstract

The Inter-university Consortium for Political and Social Research (ICPSR) is developing a machine learning approach using natural language processing (NLP) to assist in the detection of informal data references. Formal data citations that reference unique identifiers are readily discoverable; however, informal references indicating research data reuse are challenging to infer and detect. We contribute a model that uses a combination of cues, such as the presence of indicator terms and syntactical patterns, to assign a likelihood score to dataset mentions and extract candidate data citations from academic text. In production, the model will support the evaluation of candidate documents for ingest into the ICPSR Bibliography of Data-related Literature. This work supports a larger effort to measure the impact of research data.