Data Science Research Programme
Social and Behavioural Sciences
The Faculty of Social and Behavioural Sciences
The Faculty of Social and Behavioural Sciences brings together high-quality research and outstanding teaching in the disciplines of cultural anthropology, education and child studies, political science, psychology, science and technology studies, as well as in multidisciplinary approaches.
The shared essence of the Faculty’s teaching and research is that it is not pursued at some remove from society, but is instead closely and inextricably connected with it, precisely through the questions and subjects addressed. The Faculty has become a leading international centre for research and teaching in the social and behavioural sciences. The academic staff work with their students on questions relevant to larger society and often with immediate societal impact.
Data Science Research Projects
Stacked Domain Learning for multi-domain data: a new ensemble method
Wouter van Loon
The aim of this project is to develop accurate but interpretable ensemble learning methods for high-dimensional multi-domain data. Nowadays, researchers are confronted with multi-domain data more and more often. In health research, for example, multi-domain data can occur when data are collected from multiple sources (e.g. medical imaging, genomics, questionnaires), or when different feature sets are derived from a single source (e.g. different MRI modalities). Combining data from multiple domains can potentially lead to a better understanding of disease and improved early diagnosis, but it is unknown how these domains can best be combined.
The project currently focusses on the domain selection problem: How can we identify the domains that are most important for prediction? A newly developed ensemble learning method is shown to offer a large increase in both speed and accuracy compared to existing methods.
Understanding scientific progress by analysing the context of scholarly citations
The objective of this project is to fundamentally improve our understanding of the ways in which science progresses. Empirical studies have used bibliographic metadata to provide relevant insights, but these studies have failed to tell us how science progresses. Supported by computational advances and improved data access, we propose a large-scale data-driven approach in which scientific progress is studied based on the full text of scientific documents.