DataShare Spotlight: The Hiberlink project, and anticipating change in a data repository

Posted on 20 March 2025 by Evelyn Williams

While looking through DataShare for datasets about research practice, I found the Hiberlink project dataset [1], which perfectly illustrates some of the challenges the changing digital research landscape poses to good research practice.

The Hiberlink project investigated ‘reference rot’, which happens when a publication includes links that no longer contain the information the authors originally cited. The Hiberlink researchers found one in five articles in their sample contained at least one link which didn’t resolve to the originally referenced content [2].

This project garnered lots of press coverage around 2015-16, and is just as relevant 10 years on, still resurfacing on Twitter every so often. Sometimes the coverage takes a negative bent, from those who view the reproducibility crisis as proof that academia is a waste of tax money and human potential. More often though, it comes from people advocating better, more open research practices.

Richard Tobin, who deposited the Hiberlink dataset, did a good job of completing the metadata and helpfully included a link to the project’s website, hiberlink.org. However, if you follow this link it appears to have expired and been purchased by an unknown party selling “virtual data rooms”. None of us are safe.

Screenshot of the Hiberlink website having been taken over by a company selling something called virtual data rooms.

Unknown author, “What Is a Virtual Data Room? A Guide to Secure Online Document Management,” Hiberlink (https://hiberlink.org : 5 February 2025); archived at Wayback Machine (https://web.archive.org/) > https://hiberlink.org > 19 March; citing a capture dated 15 March 2025.

Why links die and why it matters

Change is a given in research projects nowadays. The path of enquiry shifts, researchers leave the university and others join, and new tools emerge for carrying out research and sharing findings. Researchers share their work on blogs, project websites, and a seemingly endless chain of social media platforms, moving with the digital tides.

But digital repositories, which are often modelled after physical archives and aim to take a static snapshot of research, can struggle to represent research projects in flux. This article takes a look at how nascent web archives used archiving frameworks taken from the field of librarianship [3].

Traceability is a core tenant of open research, but traceability can be difficult online. Digital and web-based resources often evolve over time, unlike traditionally published articles. URLs in publications or metadata may expire, or the content on the linked page may radically change. Changing content makes the research process less transparent and makes research findings harder to verify and reproduce.

Academic staff are often encouraged to have a strong online presence. Sharing research online is considered an important part of doing ‘open research’, and a strong online presence is seen as important for academics’ personal career development. But most of us don’t often think much about the web of links created during a research project. These links can drift and degrade as researchers leave an institution or move on to new projects, as a pot of funding runs dry, or as reminders to renew a domain registration end up in a spam folder.

The case of web domains is especially interesting, as they can be bought by anyone. Old links to defunct external web pages may lend credibility to questionable groups or dodgy businesses that claim a domain once a researcher’s registration expires.

Why links pose a challenge for data management

While depositing a research dataset in an open access repository seems like a natural conclusion to a project, you might be horrified to learn that the work doesn’t stop here. In data management plans we think about how we can share, store and preserve our data for the future. We don’t usually consider how all the peripheral stuff around our data might also need to be preserved and maintained.

We usually recommend that instead of using a live URL to link to a project website, researchers cite archived copies of the web pages and use persistent identifiers (like DOIs) where possible. The authors of the Hiberlink paper encourage researchers to reference archived versions of the web, like the non-profit Internet Archive. But as could have been the case with the Hiberlink project, linking to an archived copy isn’t ideal if a project website is still being actively updated and you want to direct people to the most recent version.

What can we do?

So, beyond using persistent identifiers, how can we make sure that a repository doesn’t contain a bunch of dead links? The obvious answer is that researchers should periodically review all their items in the research repositories at every university they’ve worked at, and update the metadata… forever? As is often the case with good research practice, solutions often require an ongoing investment of time that researchers simply don’t have.

Maybe as data archivists, we have to accept that metadata will decay, and that messy metadata is better than invisible research data.

Flow chart showing the data management process, with 'Update metadata' stuck inside an endless loop.

The metadata update hell loop

How can you avoid reference rot?

For help figuring out how you can preserve your websites, see the university’s resources on web archiving [4]. Dr Alice Austin does regular training sessions on web archiving, which I recently attended and found very useful. And if you’d like advice on linking all your archived research outputs together in a way that ensures they’ll be useful in years to come, get in touch with us!

Learn more about the Hiberlink project

You can read more about these issues in the archived versions of hiberlink.org from before March 2022 in the Internet Archive [5].

If you’d like to read more about reference rot, the authors of this project wrote two great papers on this phenomenon:

Scholarly Context Not Found: One in Five Articles Suffers from Reference Rot [2] looks at reference rot in over 3.5 million STEM articles from 1997-2012 and finds that one in five contains a link which doesn’t resolve to the originally referenced content.

No More 404s: Predicting Referenced Link Rot in Scholarly Articles for Pro-Active Archiving [6] explores the use of machine learning to identify links at risk of rotting.

While writing this blog post I spoke with Richard Tobin, one of the researchers on the Hiberlink project, and he had this to say: “Neither of us is working on related research now, but it would be interesting if someone were to take the data and see how much further the links have decayed.”

Links

[1] Tobin, Richard; Grover, Claire; Zhou, Ke. (2015). Hiberlink project data, 1997-2012 [dataset]. University of Edinburgh. https://doi.org/10.7488/ds/230.

[2] Klein M, Van de Sompel H, Sanderson R, Shankar H, Balakireva L, Zhou K, et al. (2014). Scholarly Context Not Found: One in Five Articles Suffers from Reference Rot. PLoS ONE 9(12): e115253. https://doi.org/10.1371/journal.pone.0115253

[3] Hegarty, K. (2022). The invention of the archived web: tracing the influence of library frameworks on web archiving infrastructure. Internet Histories, 6(4), 432–451. https://doi.org/10.1080/24701475.2022.2103988

[4] The university’s web archiving service: https://library.ed.ac.uk/heritage-collections/collections-and-search/archives/digital-archives-and-preservation/web-archiving

[5] Archived version of Hiberlink project website. https://web.archive.org/web/20220306174247/http://hiberlink.org/

[6] Ke Zhou, Claire Grover, Martin Klein, and Richard Tobin. (2015). No More 404s: Predicting Referenced Link Rot in Scholarly Articles for Pro-Active Archiving. In Proceedings of the 15th ACM/IEEE-CS Joint Conference on Digital Libraries (JCDL ’15). Association for Computing Machinery, New York, NY, USA, 233–236. https://doi.org/10.1145/2756406.2756940

DataShare spotlight: Debates on slavery and abolition held by student debating societies at the University of Edinburgh, 1765-1870

Posted on 12 March 2025 by kmunro

The best part of my job is looking through the new datasets submitted to DataShare, our open-access data repository. One of the first datasets that gripped me was; Debates on slavery and abolition held by student debating societies at the University of Edinburgh, 1765-1870.

This dataset is really cool because as well as being a valuable resource for future research projects, it’s extremely interesting to read, even as someone who doesn’t know anything about historical research. This readability is what makes humanities dataset submissions so fun to process.

This dataset summarises debates on chattel slavery and abolition by two of the University’s debating societies during roughly the last hundred official years of the Transatlantic slave trade. It includes motions and outcomes of the debates, as well as information about the people participating and the positions they took.

It’s easy to tell ourselves that people in the past caused unimaginable harm because they didn’t know any better. Maybe this impulse is a form of self-preservation, a way to deny our ancestors’ agency to protect them – and ourselves – from blame. The dataset reminds us that even at the height of the slave trade there were many people publicly voicing their opposition. The data give us some insight into how these men understood their own complicity in slavery and their responsibility in upholding or abolishing it.

It’s interesting to see, for example, that some debate outcomes were pro-abolition, but against immediate abolition. Or how a debate on whether it would be sound policy to abolish the African slave trade had a unanimously pro-abolition outcome in 1792, yet full emancipation didn’t come for over forty years.

A preview of ‘University of Edinburgh Dialectic Society debates on slavery and abolition, 1792-1870’. From Buck, Simon; Frith, Nicola; Curry, Tommy. (2024). Debates on slavery and abolition held by student debating societies at the University of Edinburgh, 1765-1870 [text]. University of Edinburgh. Institute for Advanced Studies in the Humanities. https://doi.org/10.7488/ds/7841.

The dataset comes from The Decolonised Transformations Project which aims to take a critical look at the University’s complicity and investment in slavery and colonialism, confront the legacies of these choices, and make concrete recommendations to address present-day structural racism. The project is a great example of how Humanities research can be translated into a wide array of resources to maximise its utility and reach. As well as traditional outputs like publications and reports, the research team has published datasets, done podcasts, and held workshops and talks for the wider community.

Decolonised Transformations – Confronting the University’s Legacies of Slavery and Colonialism

The datasets have been downloaded multiple times since they were shared in November, so I’m sure other people are finding this data as interesting as I am.

Evelyn Williams
Research Data Support Assistant

DataShare spotlight: Human MotionLess Dataset (HuMoLs) and the creative potential of research data

Posted on 28 January 2025 by kmunro

For the second installment of the spotlight on DataShare blog posts, I would like to showcase a fascinating item containing videos of people not doing anything!

The dataset in question is titled “Human MotionLess Dataset (HuMoLs)” and was created by researchers Longfei Chan, Muhammad Ahmed Raza and Robert Fisher, who are based in the School of Informatics’ Institute of Perception Action and Behaviour. While on the face of it, people being still might not seem very dynamic, the research behind this dataset is trying to solve a difficult problem with a very useful outcome. Simply put, how do you tell when someone is lying still because they are doing something like sleeping, or if it is because they are unwell or have fallen? The videos in this dataset aim to try and train healthcare monitoring systems to help determine whether it is the former or the latter of these possibilities, with the priority being to uncover any critical medical conditions or to analyse chronic conditions.

A selection of still images taken from the videos in the dataset.

What struck me while reviewing the videos for submission was that beyond the usefulness of these videos to the research project, there was the potential for them to be adapted creatively. The videos have a deliberately “uncanny valley” aspect, due to using AI to deepfake participant’s faces in order to preserve their anonymity. The amusingly odd character to the videos made me imagine them being used in an Adam Curtis documentary, or in an Aphex Twin music video. Possibly even in an Adam Curtis documentary with Aphex Twin music over the top of it.

This raises the fascinating idea that there are rich sources of research data stored in open access repositories that could have a life beyond being reproducible, but could also be reused, repurposed and remixed into creative new pieces, adding value to both the research itself, and to the repositories where the affiliated data is stored. To demonstrate this possibility, I have edited some of the videos together, see below, and set them to music. The piece of music, “Redolescence” by Other Lands, has a dreamlike quality to it which both complements and recontextualises the videos into something beyond their originally intended use.

What other audio and visual materials are there contained in research data repositories waiting to be repurposed in a creative manner? Time (and much more talented creators than me) will tell!

The full dataset can be found on DataShare: Human MotionLess Dataset (HuMoLs)

The paper which the dataset supports: OPPH: A Vision-Based Operator for Measuring Body Movements for Personal Healthcare

Permission to use the music featured in this video was kindly granted by Gavin Sutherland, performing here under his artist name of Other Lands. The album which contains this track can be purchased on Bandcamp: Other Lands – Riddle of the Mode

Evelyn Williams, new Research Data Support Assistant

Posted on 15 October 2024 by kmunro

Hello, readers of Edinburgh Research Data Blog!

Last month I joined the University’s Research Data Service team as a Research Data Support Assistant, and I’m excited to be back at the University after three long years working as a data scientist at tech start-ups.

Me with a bag of churros in the Montjuïc, Barcelona, where I spent a few months in the winter of 2022 – Photo credit: Evelyn Williams

This career pivot from tech into collections management feels natural to me as a lifelong collector and cataloguer. An early memory is winning a Stanley plastic small parts organiser at a village tombola, the kind you’d use to store picture hooks and screws. I’d never seen a more magical object in my life. I began hunting for groups of items tiny enough to fit in the compartments like it was my life’s work. Elastic bands, our Labrador’s fur during moulting season, glittery hair beads (it was the early 2000s), woodlice. My favourite present from last birthday was a Dymo label maker. When I first read the description for this role, working to archive the University’s research data sounded like a dream come true. It’s especially exciting to be dipping my toe into data management at a university where RDM is already so well established, thanks to the work of Robin Rice and the many others involved in developing the department and the University’s data management policy.

I’ve been curious about archives and collections for a long time. I loved interning as a Collections Assistant in Special Collections at the Sir Duncan Rice Library in 2017 while I was an undergraduate Linguistics student at the University of Aberdeen. I helped run the reading room, assisted with manuscript conservation and digitising, and carried out archive research for the Library’s exhibition. Exploring the stacks of manuscripts and ephemera, I felt like the luckiest girl in the world. The highlight of my job was getting to see a volume of Audubon’s Birds of America (1827-1838). It was an incredibly special experience for lots of reasons – the sheer size of the book (it’s a meter tall!), the beauty of the illustrations, and the depictions of bird species that are now extinct. An example of an illustration of owls is shown below.

Barn owl illustration from Audubon's Birds of America.

Audubon, J. J. (1840) Barn Owl. The birds of America, plate CLXXI. New York, J.J. Audubon; Philadelphia, J.B. Chevalier. Photo credit: The John James Audubon Center at Mill Grove, Montgomery County Audubon Collection, and Zebra Publishing.

The photo below was included in the exhibition I worked on about medical innovation in wartime. So dramatic!

A photo of a nurse tying Sir Henry Gray’s surgical mask

A nurse tying Sir Henry Gray’s surgical mask. Photo credit: George Washington Wilson & Co. (1853 – 1908). DR GRAY ROYAL INFIRMARY ABERDEEN. [Photograph]. Aberdeen: The University of Aberdeen. GB 0231 MS 3792/D0500, George Washington Wilson & Co. photographic collection.

I’m thrilled to be back at the University and working with researchers again. The last time I worked here was three years ago as a Research Assistant while doing my master’s in Speech and Language Processing, helping researchers in the Centre for Speech Technology Research to evaluate audio processing models like computer-generated voices. I learned so much by being involved in lots of different research projects, and I’m looking forward to the huge scope of people and projects I’ll support in my new role.

That role was also where I first saw the potential of open data sharing. The University’s most accessed DataShare dataset was developed and shared by colleagues at CSTR, and has since been used and cited by research teams around the world, including at Google, Deepmind, and Meta as well as at countless universities. Making this speech data publicly available has contributed to big improvements in, for example, the speech devices used by many people with Motor Neurone Disease, and in the algorithms hearing aids use to make speech clearer.

Sharing your research data may sometimes seem like an afterthought to a project, but it can have a far-reaching impact and accelerate scientific progress. My hope is that in my new role I can help to further open research in a small way.

Photo credit: The University of Toronto. (2012). Subject in AG500. The TORGO Database: Acoustic and Articulatory Speech From Speakers With Dysarthria. https://www.cs.toronto.edu/~complingweb/data/TORGO/torgo.html

This photo from the TORGO project captures the process of recording facial movement during speech using an electromagnetic articulograph machine. I used the TORGO dataset during my masters research, and I was grateful the researchers had published their data for academic use.

After I finished my masters I worked as a data scientist at a couple of tech start-ups, building artificial intelligence models. While I enjoy writing code and working on complex engineering projects, I didn’t like the restricted field of vision you have when you’re working to solve a narrow commercial problem. I’m happy to be in a more social role where I can support lots of different people and projects.

Some mugs I made for our most recent Open Studios event at Abbeymount Studios.

So far, the Research Data Service team has been really welcoming, and I feel lucky to be working with such knowledgeable and friendly people. I’ll be working 3.5 days a week with the RDS team, and on my other days I’ll likely be at the pottery studio, please see photo above, or reading. My collection of graphic novels is getting out of control, and I love fiction where nothing much happens but everything is just a bit unsettling. At the moment I’m trying to read everything by and about Shirley Jackson, as well as novels about disgruntled tech workers. Everyone I know is sick of me trying to get them to download the Libby app. (“It’s like Audible. But it’s FREE!”).

Evelyn Williams,

Research Data Support Assistant