Diving into cultural data

‘Our main research question is how to responsibly apply AI and data science in the culture and media sector. That sector is characterized by high ethical standards and clear moral principles. We translate those high level principles like inclusivity or transparency into demands for the AI pipeline. And we evaluate to what extent certain data science methods can help cultural heritage organizations achieve their goals.’

‘One of our projects is on the use of controversial terms in cultural heritage collections. Since those collection datasets may be the basis for AI systems that produce automatic descriptions or query extensions, it would be good to be able to automatically detect at a large scale which terms are controversial given the context that they are in . A second project is about evaluating the inclusivity of a recommender system for libraries, assessing for example whether or not it favors authors from a specific gender or origin.’

‘In a study on the online newspaper archive Delpher we were able to identify distinct search patterns depending on people’s interests. We discovered that people looking for WOII-related subjects use very complex search behaviour, whereas people interested in their ancestry typically have short sessions with only a couple of simple queries and very little clicks or downloads. These results lead to recommendations to digital archives on how to best support the different search behaviours of their users.’

‘Combining heterogeneous and often cross-media collections, data modelling, and semantic search. Our data consists of physical objects, short pieces of texts in natural language, newspapers, audiovisual archives… And they do not always completely meet our purpose. For example, in the book recommender project, we have data about which people read which books. But we do not have any information about the authors. A lot of our work therefore involves finding smart ways of combining datasets to distill the information we need.’

‘Both the Cultural AI Lab and the AI, Media and Democracy Lab bring together different scientific disciplines and societal organizations to collaborate for the longer term. That allows us to build a deep mutual understanding for each other’s language, challenges and expertise, which is essential if you want to make true impact.’

Laura Hollink obtained a Master of Arts in Information Science with a minor in Urban Sociology from the University of Amsterdam and a PhD in Computer Science from VU University. After holding research positions at TU Delft and VU University, she came to CWI in 2015. There she has been leading the Human-Centered Data Analytics group since the winter of 2021.