Community and Data - The Kick-Off Workshop

May 18, 2022

As already described in our previous posts, the LiviaAI-project is all about linking art collections held in three Viennese museums. What we didn’t mention is that the LiviaAI-Project is also about community and capacity building between museum, technology, and university partners.

In the conception phase of the project, we had multiple conversations with our museum partners and quickly realized that most museums had already come in contact with Artificial Intelligence in some way, but had not been able to integrate previous projects into their work and datasets. This made us curious: “why did results from previous projects not feed back into the museum? What happened with the results of previous AI-projects?” We came across two main bottlenecks: (1) the museum collections were used as data providers and the return of results to the museum was not intended. As a result, the museum partners did not know very much about the project and the processes. (2) the projects ended before the museums were able to include the results of the AI-processes into their workflows and systems. Based on these discoveries, we decided to include three goals in our project design:

1. We want to learn from the museum partners! No one understands the data better than the museum curators and data stewards.
2. We want the museums to know and understand our approach on a technical and content level so that they can include it in their collections (if they want to).
3. We want to build community between the museum partners and the LiviaAI-team.

The Workshop Format

With these goals in mind, our first and main workshop of the project took place May 3, two months following the start of the project. The workshop was intended to be collaborative, center around the knowledge of the museum partners, and help us to better understand the data provided by the museums. Eva Kleinferchner from BlueLabs in Graz supported us in the workshop design, guided us through the development of activities and led the workshop. We were very grateful for her help which and also meant that the entire LiviaAI-team was able to participate in the activities.

We invited our museum partners to attend the workshop with 2-3 staff members and in addition to the project team, we were a total of 12 participants. The group was the perfect size for an interactive setting. Overlooking Vienna from the Sky Lounge of the University of Vienna, we spent five hours poring over collection data and discussing user scenarios.

The Workshop

The workshop started off with an introduction and collaborative dive into the topic that included walking around, asking each other questions, and drawing. The perfect way to activate everyone and ensure that everyone met each other. This was followed by an overview of the project, the goals, and structure. Rainer provided an overview of the workflow from a technical perspective. Although most museum partners were aware of the overall goals of the project, we wanted to make sure that everyone knew what was going on, what we wanted to do, and what the technical steps and processes were. This led to several discussions on the scope and vision of the project as well as the future trajectory of the results.

In the next step we were split up into small teams and discussed possible stakeholders that might be involved, interested or profit from linked Viennese collections. The discussions in small groups were helpful for comparing experiences and institutional practices in dealing with cultural collections: who enters which data? Who checks the data entry? What data is open to the public? Who uses the data? Following the discussion each team settled on three main users that would be interested in accessing linked data including possible gains and pains.

Following a break for refreshments, we continued with two deep dives into the data which proved to be especially valuable to us. Rainer had prepared random data samples of ca 50 data entries from the data provided to us from the museums. The data entries were printed out on A3 and in three teams, every museum inspected their own data samples and those of the other two museums. The task was to identify commonalities between the different data strutures, or areas where one data structure might have more expansive or missing information compared to others.

Most of the participants had never seen their data in spreadsheet format and viewing the data in such a way helped think about the data and what it represents.

In comparison with the other datasets, we were able to compare how some information is represented differently depending on the museum and the museum’s mission.

In the second deep dive Rainer demonstrated the results of clustering the metadata, as described in our last blog post. Again, we provided every team with data samples of items that formed tight clusters (“close neighbors”) and a separate data sample of items that did not cluster as closely (“distant neighbors”). We wanted to know whether the close and distant records identified by the algorithm were also perceived as relevant from a curatorial and data steward perspective. We discussed why certain data entries may have clustered more closely, and on possible strategies to improve the clustering.

In a final data discussion, we compared several images of objects that had been placed by the algorithm in close neighborhoods (and thus identified as “similar”) alongside objects that had not been placed at a far distance. But I won’t get into that here since Rainer will be providing his own description of the process later this month, so stay tuned for that.

The workshop results

The workshop was extremely useful to us on multiple levels. We enjoyed finally meeting the museum partners in person after a year of virtual meetings! It was also fantastic to be able to spend time with people to discuss museums, data, and improving access.

On a content level, the museum partners really helped us to better understand their data. They pointed out certain data fields that we had either misunderstood or not fully grasped. We also identified fields where small errors had slipped into the data export.

Most importantly though, the museum partners helped us to better filter the datasets and improve the parameters we are using to cluster the object data. We now have a much better idea on how to proceed with the project.

Looking back at the workshop I would like to emphasize one last take away: the challenges of digitizing collections while also managing technological advances in a rapidly changing world is not something that can be done alone. Besides, it’s a lot more fun, especially when you find a group of people willing to discuss museums, data, and technology.