Leonard&Dhollander_OpenScienceBelgium.pptx
•Download as PPTX, PDF•
0 likes•21 views
Presentation at the “Open Science: connecting the actors” event on the 21st of November 2022: Share best practices, foster community, and encourage knowledge-sharing on Open Science. At the heart of the Open Access Belgium community is the ambition to open up the way we organize and conduct scientific research. The Open Science teams of the Belgian universities have developed and tested a wide range of training methods, training materials, networking activities and data solutions to facilitate and foster Open Science. Achievements, tools and lessons learned by different institutions will be shared in this networking event. Programme can be found here: https://openaccess.be/2022/10/04/open-science-connecting-the-actors/
Recommended
More Related Content
Similar to Leonard&Dhollander_OpenScienceBelgium.pptx
Similar to Leonard&Dhollander_OpenScienceBelgium.pptx (20)
More from OpenAccessBelgium
More from OpenAccessBelgium (20)
Recently uploaded
Recently uploaded (20)
Leonard&Dhollander_OpenScienceBelgium.pptx
- 1. POST-INGEST CURATION: CURATING WITHOUT AN INSTITUTIONAL REPOSITORY Kevin Leonard - Evelien Dhollander
- 2. Data Curation Requires Datasets • Data curation: adding value to (meta)data for long-term preservation • Imagined (ideal) workflow: 1. Researcher provides data to curator for curation • Voluntary submission • Automatic part of ingest in institutional repository 2. Curator makes changes and recommendations 3. Data is put online for long-term preservation • Is that realistic for many institutions?
- 3. What Often Happens to Datasets, Really? Scientist Curator + + + +
- 4. Proposed Workflow 1. Find datasets online • Employ existing data linking architectures • Use repository APIs 2. Produce (meta)data augmentation plan for discovered datasets • Develop plan based on current best practices for FAIR metadata • Recommend changes that maintain existing DOI networks 3. Provide researchers with an easily actionable curation plan
- 5. Step 1: Where Are The Datasets? • Difficulties: • Datasets are broadly distributed • Affiliation information is not located in a consistent location (or format!) • Existing data linking systems (e.g., Scholix, DataCite) have limited coverage • Solution: • Use repository APIs to search for institutional datasets • Search outside of just <creator><affiliation> field
- 6. Example Python Code • Python code to search for institutional records • searchQuery can include multiple items • Universiteit Gent • UGent • Ghent University • 00cv9y106 (ROR id) • Saves DOIs of all datasets to csv • Can use OAI-PMH to extract more metadata information • Focused on several popular repositories, easily extended • Zenodo, OSF, Dryad, Figshare, PANGAEA
- 7. Step 2: What To Do With What You’ve Found • Repositories often allow metadata fields to be edited • WITHOUT triggering the creation of a new version (and therefore a new DOI) • Editable fields vary by repository: STRI CT LENIEN T • Editing any metadata fields creates a new version • Most fields can be edited • Title, authors, relatedTO
- 8. Develop Recommendation Plan • Is the title clear? • Are keywords provided? • Are there links to related publications? • Do the authors have linked ORCIDs or affiliations? • Is there sufficient documentation in a README? • Can this information be provided in <description descriptionType=“Abstract”>?
- 9. Step 3: Communicating the Recommendations • Implementation relies on participation of the researcher • Curation plan must be easily actionable with clearly articulated benefits • Reduce burden on researcher to interpret instructions Metadata Field Current Value Recommended Changes Rationale Title Abstract …
- 10. Current results • Currently, the code harvests >2000 total records • Frequently encountered issues: • Abstracts redundant with publication • No direct contact information • Missing keywords Source Number of Records Found DataCite 236 Dryad 196 Figshare 302 OSF 186 Pangaea 724 Zenodo 710
- 11. Conclusions • Relatively simple method to provide value to existing datasets • Benefits even if author declines to make recommended edits: • Helps institution find their research outputs • Provides researchers with FAIRness recommendations that they can implement for future datasets • Communicates the existence (and utility!) of data support staff
- 12. More information? Contact us: Kevin Leonard kevinmichael.leonard@ugent.be Evelien Dhollander evelien.dhollander@ugent.be
Editor's Notes
- Thank you. Today we are going to talk about post-ingest curation, and what we at UGent have been considering to curate research data despite not having an institutional repository.
- As I’m sure you are all aware, research institutions are becoming increasingly aware of the importance and utility of data curation, which, for our purposes, we can broadly define as any activity that adds value to data or metadata prior to its long-term preservation in a data repository. It is usually conceptualized as an ideal workflow, wherein the researcher provides their data to a curator for curation. This can either be a voluntary submission, as envisioned here in this diagram from the Data Curation Network, with the researcher actively seeking out and requesting the assistance of a curator, or it can be automatic, such as when a researcher deposits their data in an institutional repository and that institution’s curators can immediately begin to work on the data, making the curation a necessary part of the data’s path towards preservation. Regardless, the curator is then able to make changes and recommendations, ideally through some kind of back and forth dialogue with the researcher, before finally the curated data is put online for long-term preservation. Therefore, in this conception, curators always get to curate the data BEFORE it is published online. What we asked in this project is how realistic that workflow is for most researchers and institutions, and whether an alternate model might be necessary for cases in which such curation is not so automatic.
- Looking outside the ideal, we turned our focus on what often actually happens to datasets in the research data lifecycle. The scientist completes some research project, generating a manuscript for publication and some associated data. They submit their manuscript to a journal and (as it approaches acceptance) have to publish their data online. Even if the researcher knows about curation services available to them at their institution, they might not feel that they have time to go through rounds of curation as they need their datasets online NOW, and so they circumvent the data curators and deposit the datasets directly into the general or domain-specific repository of their choice. They annotate the dataset with metadata according to their own understanding of best practices and what little time they have available to dedicate to documentation, and the data then sits online in the repository without ever having the opportunity to have value added to it by data curation specialists. Note that, while this workflow is possible for researchers from any institution, it’s especially likely for institutions that don’t have their own institutional repository, as the data curators will never have datasets automatically pass through their desks on the way to the institutional repository. What we aimed to do in this project is to define an alternate workflow for curators, wherein they can go out and find these datasets where they are posted online. Then, once they have knowledge of these datasets that are associated with their institution, they can develop individualized recommendation plans for the creators of those datasets, with the hope that the researcher implements those changes, thereby improving the FAIRness of those datasets.
- Our proposed workflow comes in three steps: First, we find the datasets that have already been posted online. To do so, we first looked at existing data linking architectures, such as Scholix, but ultimately ended up relying on repository APIs for the most popular repositories for researchers from our institution. Then, once we’ve found the datasets through these various methods, we can develop augmentation plans for these uncovered datasets. This is because many popular repositories actually allow users to edit the metadata of their published datasets without triggering the generation of a new version, therefore preserving existing DOI link networks. So, it should not be considered “too late” to curate a dataset just because it has already been hosted online. There are still things that can be done to improve its FAIRness. Finally, once we’ve developed a set of recommendations for a given dataset in an online repository, the last step is to create an action plan that can be communicated to the researcher, providing them with an easily actionable way to improve the FAIRness of their own datasets.
- So, the first step is to find the datasets online. This is more difficult than it sounds, because datasets are broadly distributed across many different repositories. To make matters worse, the affiliation information is not consistent, in location or format. Some records have the affiliation information associated with the creators. Some use the name of the institution written out in full, whereas others use the ROR, a specific id for institutions. For these and other complicated reasons, existing data linking systems end up missing a lot of the datasets that are out there. This can be easily verified… If you compare the results from using these services to just going onto one of these repository pages and entering the name of your institution into the search bar, you’ll find many records that these systems fail to pick up. Our solution then is to use the APIs to find as many additional institutional datasets as possible, and wherever possible, by searching outside of the CREATOR:AFFILIATION field.
- We’ve written some python code which harnesses the APIs of popular repositories to search for institutional datasets. Importantly for us, and probably for many institutions in Belgium, is our institution is known by many names, all of which we see authors freely use when tagging their datasets. As currently implemented, the code saves the DOIs of all the datasets it finds to a csv because that is most important to ingest into the systems that we use, but you could easily use OAI-PMH or alternate systems to extract more metadata. Lastly, we focused on the main repositories which are used by researchers from Ghent University, but this could easily be extended to focus on other repositories, insofar as they have APIs to plug into.
- Once the dataset records have been located, the next step is to figure out what to do with what you’ve found. The first part of that is determining what you can edit without triggering the creation of a new version (and therefore a new DOI). Even though these new DOIs are typically linked to the DOIs of the older versions, our thought was that it is avoid these potential issues. Different repositories vary with respect to which metadata fields are editable without triggering a new version, from very strict repositories (like Dryad) which allow essentially no editing, to very lenient repositories like Zenodo, for which you can edit almost anything, including the title, abstract, and authors.
- Once you’ve decided which metadata fields are in principle editable, you can then develop an individualized recommendation plan for that record. What exact recommendations you provide will depend on your institution’s priorities, current best practices, but we’ve collected here a few of the major items that could be included in such a plan: is the title clear? Are there keywords? Has it been linked to a publication? Are there ORCIDs linked? Have the authors provided something like an ROR? Did they provide a detailed README, and if not, could that information be provided in the abstract field?
- The last step is to communicate the recommendation plan to the researcher. Because the actual implementation of the plan relies on the participation of the researcher, steps should be taken to maximize the likelihood that they cooperate. For this, we envision a clearly articulated plan like in the table shown here, which outlines the metadata field in question, what that value currently contains, what the curators believe should be changed for that field, and their rationale. Anything that can reduce the burden on the researcher and lets them clearly see the reasoning and benefit behind the recommendations.
- This is all very interesting off course but as the saying goes: ‘the proof of the pudding is in the eating’, so here are some of our results. By running the code we gathered over 2000 dataset records from five major repositories and DataCite. We analysed a subset of these records to get some idea of the issues we will encounter in the future. A first issue is the redundancy of abstracts: most datasets have the same abstract as their corresponding publication. This isn’t necessarily a big issue when a datasheet or README is provided for the dataset but when the abstract is the only information on the content of the dataset, not enough information might be present for researchers to reuse the data. A possible solution to thids was mentioned a few slides ago: a README could be provided in the abstract metadata field. A second issue is the absence of contact information in the dataset metadata. In most cases the contact information is found via the linked publication, so contact information can be found but this is not a good practice, we want to encourage researchers to provide contact information in their dataset metadata as well. A third issue concerns keywords as in: there are no keywords provided. The dataset should at least have some of the related publication’s keywords and ideally have its own specific keywords to improve FAIRness.
- This is the basis of our proposal to provide curatorial benefits after a dataset has been already uploaded to a repository. Of course, we don’t want to suggest that this solves all problems, and it will not find ALL datasets, but it still has several benefits. Even if the author declines to make the recommended edits, the integration with repository APIs helps institutions find their research outputs. And the emails to the researchers, which include the detailed plan of how to improve the FAIRness of their datasets, provides the researchers with knowledge that they can carry with them in the future, and works as a way to let them know of the curatorial services that your institution might offer, and how they can help improve their online data.
- If you would like more information, please contact us. Thank you for your attention!