Abstract
Data have become central to scientific discovery. While primary data collection remains vital, there is growing recognition of the benefits of reusing existing datasets. However, identifying suitable datasets for specific research questions is increasingly difficult due to the fragmentation and heterogeneity of the big data ecosystem. Despite the expansion of data sharing, efficient dataset discovery remains elusive, with limited empirical research on how datasets are identified, interpreted, and reused. Current dataset search practices often lack standardization, leading researchers to rely on convenience rather than systematic criteria. Unlike bibliographic research, dataset selection lacks a formal methodology, increasing the risks of bias, inefficiencies, and reduced generalizability. To address this gap, we introduce datagraphy, a structured approach to dataset identification and evaluation. Analogous to bibliographic methods but designed for datasets, datagraphy encompasses not only discovery but also critical assessment of dataset quality, relevance, interoperability, completeness, sustainability, and ethical use. By formalizing dataset search as a research practice, datagraphy seeks to improve transparency, reproducibility, and interdisciplinary collaboration, while also reducing research redundancy and environmental impact. We present a 9-step framework to operationalize datagraphy and explore challenges such as inconsistent metadata and variability among dataset discovery tools. This framework provides a foundation for systematically and reproducibly identifying and synthesizing reusable datasets. To demonstrate the application of the proposed framework, we conducted a datagraphic search focused on the exposome. We discuss major challenges faced by datagraphy with respect to metadata availability, repository heterogeneity, dataset accessibility, and dataset quality, as well as highlight how datagraphy could enhance transparency, reproducibility, and efficiency at the researcher level. Datagraphy is intended to complement repository-level improvements. Aligning researcher practices with standardized, machine-readable metadata, persistent identifiers, artificial intelligence integration, and lightweight packaging frameworks such as RO-Crates and FAIR (Findable, Accessible, Interoperable, and Reusable) Digital Objects could enable automated discovery and sustainable dataset reuse. By integrating structured researcher-level methodology with systemic improvements and community efforts, datagraphy could offer a scalable approach for systematic, FAIR-aligned data-driven research across disciplines.