The future of data citation: ideas from IASSIST 2017

IASSIST 2017 took place at the University of Kansas in Lawrence from 23 to 26 May, under the theme “Data in the Middle: The Common Language of Research.” Its archived programme, schedule and plenaries remain valuable evidence of how research communication was being reconsidered when big data and deep learning were already changing academic practice. The future of data citation must build from that shared language rather than treat data as a technical side issue.

The programme connected work on digital agriculture, global food security, large-scale information systems and research communication. These fields exposed a practical problem: a dataset can be rich, licensed and technically available, yet still be difficult to credit, verify or reuse. Strong data citation links a research object to a persistent record, identifies the people and institutions responsible for it, and preserves the context needed to interpret it responsibly.

For an Australian audience, that responsibility includes familiar pressures and distinctive choices. Universities, agencies, farms, businesses and public repositories operate within Commonwealth and state requirements, while communities expect research to respect privacy, cultural authority and commercial confidentiality. Citation practice therefore needs to work across sectors, recognise Indigenous data governance and support transparent machine-actionable metadata without pretending that every dataset should be open.

Earlier emphasis	Shared-research challenge	Future-oriented practice
A stable reference for a dataset	Identifying which release was analysed	Versioned identifiers linked to maintained records with creators, dates and stewardship details
Authorship on the paper	Recognising data collectors, curators, developers and domain experts	Linked contributor roles and shared credit across papers, datasets and software
A description in an appendix	Reproducing transformations across different systems	Machine-readable provenance, methods, code dependencies and validation checks
Open access as the default goal	Reconciling privacy, cultural and commercial expectations	Clear licences, defined access categories and reasons for restricted or mediated release
A final PDF as the research output	Deep learning, simulation and frequently updated data	Citations for model versions, input data, workflows and meaningful output records

From identifiers to credit

At IASSIST 2017, data citation was not simply a question of adding another reference to a bibliography. It concerned whether researchers could exchange evidence across disciplinary and institutional boundaries. The common language of research worked only when participants could recognise the object behind a claim and follow its history.

Earlier citation systems concentrated on journal articles and final datasets. Complex projects now produce several connected objects: cleaned extracts, annotated images, code, trained models, simulation configurations and frequently updated sensor streams. Each may require a different level of access, but each can still receive a stable citation. A Digital Object Identifier is useful only when it resolves to a maintained record that explains what was cited.

Credit is the larger shift. A dataset may depend on field technicians, translators, software developers, community partners and data stewards whose names rarely appear in the paper’s author list. A useful future system treats those contributions as traceable parts of the scholarly record. It connects credit for intellectual contribution with responsibility for stewardship, documentation and long-term access.

Machine-actionable metadata and research workflows

The case for data citation becomes stronger when a publication is read by software as well as by people. Machine-actionable metadata lets a repository service expose creators, dates, licences, access conditions, methods and relationships in a consistent form. That consistency supports discovery across institutional repositories and national services, rather than relying on a person to interpret inconsistent notes in a readme file.

Workflow provenance records how a raw observation became an analytical input. It can link source identifiers to filtering, transformations, code versions and the resulting figure. For repeated studies, this trail makes sensitivity checks and replication more feasible. It also prevents a revised input from silently leaving the citation attached to an earlier result.

Deep learning adds a specific citation need: the contribution is often a pipeline rather than a single artefact. Researchers may cite a released model, the training or input data, the software environment and the evaluation set. A citation that names only the published weights can conceal the material conditions behind the performance claim. Recording data lineage and evaluation conditions makes a stronger claim of reproducibility possible.

Digital agriculture and global food security

Digital agriculture turns sensor readings, satellite imagery, weather records and historical yields into a chain of models and decisions. A citation to a final harvest forecast is incomplete if the underlying observation period, preprocessing and model version cannot be found. Field validation also matters because an algorithm trained in one soil, rainfall or farming system may perform differently elsewhere.

Global food security research depends on comparable evidence across countries, scales and disciplines. Harmonised definitions, geographic boundaries and uncertainty notes often determine whether a dataset can support a policy claim. Citation practices should expose these choices, allowing partners to assess a figure without assuming that a shared label represents identical measurement.

Australian projects in cropping, pastoral management, water planning and environmental monitoring operate in a demanding local market. Teams need outputs that can move from a university repository to a government agency or industry partner without losing their context. In Melbourne and Sydney, research groups often span universities, hospitals, councils and industry, making connected institutional identifiers more useful than a citation tied to one employer’s filing system.

Privacy, legislation and Indigenous authority

Australian research data operates within a mixed framework. The Australian Privacy Act 1988 and the Australian Privacy Principles apply to many organisations handling personal information, with oversight and guidance from the Office of the Australian Information Commissioner. Citation records can expose sensitive fields, so access labels and disclosure decisions need to reflect applicable obligations instead of being treated as publishing formalities.

De-identification is not a guarantee of safety, particularly when small communities, precise locations or rare events allow re-identification. A metadata-first service can describe a dataset, name its custodian and explain the permitted access process while keeping files closed. That approach gives credit without unnecessarily publishing personal or commercially sensitive material.

Indigenous authority adds a further dimension. The CARE Principles for Indigenous Data Governance complement the FAIR principles of findability, accessibility, interoperability and reuse by placing collective benefit, authority to control, responsibility and ethics alongside technical description. Australian First Nations communities may set their own conditions through research agreements and community protocols. A citation service should record those conditions accurately and never treat openness as permission to ignore them. Legal compliance and individual consent do not automatically settle cultural appropriateness or collective authority.

Reproducibility, credit and durable stewardship

Citation becomes part of reproducibility when it records enough detail to reconstruct the evidentiary path. That includes sampling boundaries, calibration, cleaning decisions, software dependencies and the relation between inputs and outputs. It does not require every analyst to rerun the entire project, but it allows a reader to identify what can be checked and what remains an assumption.

Credit needs comparable persistence. If a dataset is cited but a person is named only on a temporary project website, the scholarly record will not preserve the contribution. Linked contributor identifiers, role statements and funding acknowledgements help institutions recognise data work alongside conventional publications. These records also make software and model citations easier to maintain as teams move between employers.

Durability depends on governance as much as technology. A DOI can identify a landing page, yet it cannot guarantee that the files, documentation or curation staff will remain available. National and institutional repositories need clear succession arrangements, version policies and preservation commitments. Restricted data deserve citations too, using a stable record and explicit access conditions rather than disappearing from the literature because public download is impossible.

Practical measures for responsible data citation

IASSIST 2017’s programme illustrates the value of bringing data specialists and domain researchers into the same room. A contemporary programme can learn further through a speaker survey that tests whether proposed sessions match what attendees expect from evidence, reuse and responsible access. The method is practical: it connects programme design with the knowledge needs of participants.

Australian implementation should reflect everyday research habits. Teams commonly move between a shared drive, a cloud platform, analysis code and a repository submission, sometimes across several institutions, while many use public cloud services alongside university-managed systems. The citation should describe this real chain, including service changes and retained versions, rather than reward only a clean final upload.

The local research market already offers useful building blocks, including the Australian Research Data Commons, the Australian Data Archive, Trove and institutional repositories. Interoperability is more valuable than another isolated service, because Australian funders, universities and public agencies increasingly need to see connected objects, contributor roles and access decisions across systems.

Signals of a trustworthy citation record

• Every cited dataset, code release and model has a persistent identifier that resolves to a maintained landing page.
• Versions, creators, collection methods and software dependencies provide enough provenance to explain how a result was produced.
• Access conditions distinguish open delivery, mediated access and embargoes, including the reasons a record cannot be downloaded.
• Contributor roles and funding acknowledgements connect data citations with the scholarly record and recognised community authorities.

Measures for evaluators and funders

• Citation quality appears in assessment and grant reporting as a contribution to reusable research, rather than as a count of deposited files.
• Repository selections follow a common core metadata profile while allowing domain-specific fields for geography, taxa, instruments and uncertainty.
• Researchers receive linked credit for datasets, software and methods, with stable stewardship arrangements after a grant ends.
• Evaluation combines link resolution, metadata completeness, appropriate access decisions and evidence of documented reuse.

For Australian institutions, the most credible path is to make good citation part of ordinary research support, not an optional extra at submission. The future of data citation will be secure when credit, context and controlled access travel together with every research object.