Awards for Data Science projects 2017

A commision consisting of experts of Digital Humanities, Data Science & Systems Complexity (DSSC), the eScience Center and representatives of the CIT has granted six data science proposals this year.

The researchers of the granted projects get support from data scientist of the CIT to a maximum of 450 hours. The commision received twelve proposals in total. The call for proposals is being financed from the ICT strategy plan 2016-2020 of the RUG. The next round of the call of proposals for data science will be in 2018.

An overview of the granted projects for 2017. For the original version please visit the CIT website.

SPRINGH – Sensor data Processed for Reliable Innovative Health Guidance

Prof.dr.ir. J.C. Wortmann, Faculty of Economics and Business
Employers use Health Promotion Plans nowadays to provide wearables to employees, with the purpose to promote employee health self-management. These wearables provide feedback to employees. Currently, the effect of this feedback is not clear. Therefore, it is investigated in field experiments within the project SPRINT@Work, which is a collaboration of three faculties (UMCG/Medical, Social Science, and Economics and Business) and nearly 20 business partners. In this project, sensor information is combined to provide enriched feedback from various sensors to users. The effect of this feedback is currently being investigated. The next step after the project SPRINT@Work is to shift from short term effects to long term effects. However, the data science challenges of these long term effects are substantial: they include all kinds of machine learning techniques in order to select the best rules for generating feedback.

Platform Pop: Spotify’s Role as an Intermediary in the Music Industry

Dr. R. Prey, dr. M. Esteve Del Valle, Faculteit of Arts, Centre for Media and Journalism Studies
Spotify and its competitor music streaming services have become important intermediaries in the music industry. There has been much speculation in academic and industry circles over what this shift in music consumption means for the distribution of power between the major record labels, independent labels, and unsigned artists. To-date there has been no in-depth empirical research on how streaming platforms promote music, and their role in either consolidating or democratizing traditional power relations in the music industry. This study will attempt to investigate the types of artists and playlists that Spotify promotes through Twitter, and how they promote them. It will employ machine learning to analyze the content of the tweets of the Spotify Twitter account and a statistical analysis to understand what type of playlists Spotify promotes and how these playlists are promoted. It will combine this with a social network analysis of the intensity of the communication flows between the Spotify’s Twitter account and the major record labels or independent artists.

R ewriting Nature: studying the early modern scientific vocabulary in the Republic of Letters with word2vec

Dr. A. Sangiacomo, Faculteit Philosophy, History of Philosophy
Studying the transformation of early modern natural philosophy in its actual historical complexity is a major challenge for historians of philosophy and science. However, the size of the historical corpus that should be taken into account to fully address this challenge is hardly treatable with traditional close reading of a few selected texts. In this project, we plan to adapt a word2vec tool, ShiCo, to study the transformation of the conceptual vocabulary of natural philosophy across time. We plan to adapt ShiCo to work on an early modern corpus of learned correspondences (available through the Oxford E-Enlightenment project). ShiCo will allow historians to uncover the patterns in the dramatic transformation that the discipline underwent during the period. We will rework ShiCo in order to introduce a metadata filtering system capable of distinguishing different sub-classes of data and explore what is the minimum size of data sets with which ShiCo can work. This will allow ShiCo to be customized for the specific kind of corpora often studied by historians of philosophy and science. Moreover, we will experiment how Shico can deal with the kind of relatively rare technical terminology that is most relevant for the philosophical discussion. This highly interdisciplinary project will offer a platform to integrate computational approaches in the field of history of science and philosophy.

Central Sensitization and Physical Activity in patients with Chronic Low Back Pain; Exploration of movement patterns using new analytic strategies

Prof.dr. M.F. Reneman, Dr. R. Dekker, Dr. H.R. Schiphorst Preuper, Dr. I. Stuive, MSc. J. Ansuategui Echeita, Dr. C.J.C. Lamoth, Faculty of Medical Sciences/UMCG, Department of Rehabilitation Medicine
Chronic low back pain has a high prevalence and high disability impact on the individual and society. Many patients with chronic low back pain develop a dysregulation of the nervous system, called Central Sensitization. Because Central Sensitization includes alterations in the sensory and motor nervous systems, changes in Physical Activity patterns are plausible. Evidence on Physical Activity in Chronic Lower Back Pain, however, is conflicting and research immature. In this research, data of one hundred people with lower back pain is collected by means of an accelerometer. With this data the association of Physical Activity patterns and Central Sensitization in patients with Chronic Lower Back Pain will be explored. Moreover, we will analyze whether a change in Central Sensitization in patients with Chronic Lower Back Pain is related to a change in Physical Activity patterns. Finally, differences in Physical Activity between responders and non-responders to therapy will be analyzed.

Automatic recognition of Frisian speakers: using computers to discriminate the Frisian accent and voice

Dr. M.B. Wieling, Dr. N.H. Hilton, Faculty of Arts, Computational semantics / Frisian Language and Literature
Accents have been the focus of scientific interest for the insight they provide into second language learning and language contact, but have also been extensively studied from a sociolinguistic perspective to understand the performative powers of language. Questions that remain largely unsolved are what properties determine an accent exactly, how an accent is related to speakers’ language proficiency, and to which extent accentual properties are identifiable by humans on the one hand, and computers on the other. In this project we will manually annotate different accents for different words in the Frisian language. Subsequently, we will use machine learning techniques like support vector machines to cluster the different Frisian pronunciations of these words. The end result will be a map with the different accents of Frysian in Friesland. This will improve our knowledge about the different accents in the Frisian language.

Automatic Processing of Eye Tracking Movies

Dr. J.C. van Rij-Tange, prof.dr. J. Järvikivi, University of Alberta, Department of Linguistics
Not much is known of how children use and process referring expressions outside the lab. Therefore, we would like to use the eye tracking glasses for recording of child speech and child directed speech, while interacting with a picture book. Collecting a corpus like this is crucially dependent on automatic annotation, because it will take too much time and money to transcribe the data by hand. Therefore we will develop a software tool for automatic annotation of data collected with eye tracking glasses. The tool will use a pre-trained deep learning network for automatic detection and recognition of objects in the participant's visual field. It will also use a deep learning network for automatic speech to text conversion. Finally, the linguistic information is linked to the visual scene and the gaze data. Automatic processing of the data will enable us to build a corpus that combines child (directed) speech with video and gaze data. The resulting software tool will be made available for other researchers, because it is not only applicable for data collected with eye tracking glasses, but also for video data.