Turning conference presentations into citable research datasets

Conference presentations often contain valuable research evidence that never appears in a journal article. Slides may show a model architecture, a fieldwork map, a survey instrument, a data dictionary, or results that point towards a later publication. Once the event ends, however, those materials can become difficult to discover, interpret and cite.

The archived IASSIST 2017 website offers a useful setting for understanding this problem. The University of Kansas conference, held in Lawrence from 23–26 May 2017, brought together discussions about data in research, including big data, deep learning, digital agriculture and global food security. Its programme, plenary information and presentation materials demonstrate how much scholarly context can sit outside conventional publications.

Turning those materials into a citable dataset means doing more than saving a PDF. The work involves identifying the source, extracting meaningful research objects, documenting their provenance, checking rights and creating a persistent record. For Australian researchers, librarians and data stewards, the process can align with Australian Research Data Commons practices, institutional repositories and the expectations of funders that research outputs remain findable and reusable.

Start with the presentation as a research source

A presentation should be treated as a compound research object rather than a single file. The slide deck may be accompanied by speaker notes, an abstract, a conference programme entry, a recording, supplementary code or a poster. Each component can have a different creator, licence, date and level of evidential authority.

Begin by capturing the conference context. Record the event name, host institution, location, dates, session title, presenter names, affiliations and the original web address. For IASSIST 2017, this includes the conference identity, the University of Kansas setting in Lawrence and the relevant session or plenary details. A file named slides.pdf has little long-term value; a record linked to a particular session and author is much easier to verify.

The next step is to define what the dataset will contain. It might be a structured catalogue of presentations, a collection of extracted research claims, a dataset of figures and captions, or machine-readable metadata about methods and topics. These are different products. A catalogue can be created from public programme pages, while a dataset of extracted figures may require permission from presenters and careful review of image quality and context.

This distinction matters when the source is an archive. An archived conference website may preserve the programme but omit files, alter links or present pages in a way that differs from the live event. Keep a snapshot of the source page, note access dates and identify whether a file was supplied by the organisers, a presenter or an external repository.

Extract structured information from slides and programmes

A reliable workflow separates extraction from interpretation. First, preserve the original file in its downloaded form. Then create a working copy for optical character recognition, text extraction or manual transcription. Never replace the original with an OCR-processed version, because errors in names, numbers, equations and technical terms can be difficult to detect later.

Useful fields include presentation title, presenter, organisation, session, abstract, keywords, slide number, text excerpt, figure caption, data source, method, geographic scope and links to related research. For a presentation on digital agriculture, for example, “remote sensing” or “yield prediction” may be useful keywords, while a statement about a specific crop, region or sensor should be stored as a more precise claim.

A data dictionary should explain every field. If geographic_scope contains values such as “global”, “Kenya” and “Murray–Darling Basin”, document the permitted values and how ambiguous cases are handled. If a slide says “accuracy improved by 12%”, record whether that is a percentage point increase, a relative increase or simply the wording used by the presenter. Do not silently convert an informal presentation claim into a definitive research finding.

Automated tools can accelerate the process, especially across a large conference archive. Text extraction can locate repeated terms, while image recognition can identify charts and maps. Human review remains essential for citations, mathematical notation, tables and visual material. A sensible audit record includes the extraction tool, version, date, file hash and reviewer decisions. This gives users a transparent path from the original presentation to the structured record.

Build provenance, identifiers and citation metadata

A citable dataset needs provenance that another person can follow. At minimum, document who created the dataset, which source materials were used, when extraction occurred, what transformations were applied and which version is being cited. Where possible, retain the original presentation alongside the derived files, a README, a data dictionary and a changelog.

Persistent identifiers make the record durable. A DOI from a repository is generally preferable to a changing conference URL, while an ORCID can distinguish presenters with similar names. Repository metadata should include creators, title, publication year, publisher, resource type, subject terms, language, version and related identifiers. The original presentation can be linked using a “is derived from” or “is referenced by” relationship, depending on the repository’s schema.

Citation style should reflect the object being cited. A citation for a presentation may name the speaker and conference, whereas a citation for an extracted dataset should name the dataset creators and version. For example, a record might identify “IASSIST 2017 presentation metadata, version 1.0” as the cited object and separately provide the title, presenter and session of the source presentation. This prevents users from citing an entire archive when they actually relied on one extracted record.

Australian institutions commonly use repositories connected to the Australian Research Data Commons ecosystem, university data platforms or discipline-specific services. A project team in Brisbane, Perth or Canberra may have access to different repository arrangements, so the preservation plan should be agreed with the university library or research office early. A DOI alone is not a preservation strategy: the repository must also support versioning, backups, metadata export and long-term access.

For an archived event, include both the event date and the dataset publication date. The presentation may date from May 2017, while the curated dataset could be released in 2025. Those dates describe different stages of the record and help users understand whether they are citing the historical source or the later scholarly intervention.

Resolve copyright, consent and ethical limits

Public availability does not automatically mean unrestricted reuse. A presentation hosted on a conference website may contain copyrighted photographs, commercial maps, third-party datasets, unpublished results or personal information. The conference organiser may have permission to display the slides without having permission to redistribute every embedded object in a new dataset.

Classify material before publishing it. Public-domain or openly licensed text may be reusable with attribution, while a figure reproduced from a journal article may need to be linked rather than copied. For restricted content, publish descriptive metadata, a rights statement and a pointer to the original source instead of placing the file in an open download. Keep a record of permissions and decisions so that future curators do not have to repeat the investigation.

Ethical review is equally important. A slide can reveal a participant group, a small community, a farm location or an individual’s quotation even when the original study was approved for a narrower purpose. Location data from agriculture and environmental research may create risks when released at fine resolution. Aggregation, redaction or controlled access may be necessary, particularly where Indigenous knowledge, community data or culturally sensitive information is involved.

Australian practice needs to account for Indigenous Data Governance and the expectations of Aboriginal and Torres Strait Islander communities. A presentation about land, health, biodiversity or cultural materials should not be treated as ordinary content simply because it appeared at a public conference. Consultation, appropriate attribution and community authority may affect whether the extracted material can be openly deposited.

Use a clear rights field for every item: open reuse, attribution required, permission pending, restricted access, or link-only. Add a plain-language note for users. Phrases such as “available online” and “free to read” do not tell a future researcher whether a chart can be copied, modified or redistributed.

Publish a useful package rather than a loose file

The final deposit should be understandable without the curator standing beside it. Include a README that describes the project purpose, source archive, inclusion criteria, extraction process, quality checks, known gaps and recommended citation. Add the data dictionary, licence or rights statement, file list and a changelog. If the collection includes images or transcripts, explain their relationship to the metadata.

A practical package might contain a CSV file for simple discovery, JSON for structured exchange, a human-readable catalogue and a folder of permitted source files. Use stable field names and UTF-8 encoding. Preserve accents and non-English names accurately, and distinguish missing, not applicable and not yet checked values. These small decisions make the collection far more useful to researchers who want to analyse it later.

Quality assurance should include duplicate detection, name checks, broken-link testing and comparison against the original conference programme. In Australia, a project team working across Sydney, Adelaide and regional campuses may also need to account for different local repository systems and staff turnover. A handover document reduces the risk that a valuable collection becomes dependent on one person’s laptop or institutional login.

The audience for the dataset should shape its metadata. A digital humanities researcher may want full-text excerpts and topic terms, while a research data librarian may need provenance and rights fields. Someone studying the history of machine learning might search by method, year and discipline. Good discovery metadata supports all three without claiming that extracted presentation content has the same status as peer-reviewed results.

The comparison below shows how common outputs differ in durability and citation value.

Output	What it preserves	Main limitation	Recommended citation use
Original slide PDF	Visual layout, wording and figures as presented	Often lacks structured metadata and may contain inaccessible text	Cite when discussing the presentation itself
Conference programme record	Event, session, presenter and schedule context	Usually contains little research detail	Cite for event and session information
Extracted metadata dataset	Searchable titles, people, topics, methods and links	Depends on curator decisions and source completeness	Cite the dataset version used for analysis
Figure or claim catalogue	Selected visuals, captions, claims and provenance	Requires careful interpretation and rights review	Cite a specific record or version
Curated research package	Source files, metadata, documentation and transformations	Takes the most effort to maintain	Cite as the primary reusable research object

Versioning is important when a record changes. A corrected presenter name, newly granted permission or repaired extraction should produce a new version rather than silently overwriting the old one. The latest version can be recommended for general use, while earlier versions remain available for reproducibility. Use semantic versioning or a repository’s built-in version system and explain the changes in plain English.

A deposit can then be promoted through the original conference archive, institutional research profiles, library discovery services and subject networks. Australian researchers often need outputs to be visible to both local colleagues and international collaborators, so rich metadata and an English-language abstract are valuable even when the underlying research concerns a particular region. The result is a durable scholarly record: traceable to the conference, transparent about its limits and ready to be cited in future work.