For best experience please turn on javascript and use a modern browser!

Computer scientist Marcel Worring was ahead of his peers when he decided to investigate the ‘Shape Analysis of Digital Curves’ for his PhD at the University of Amsterdam (UvA) in the late eighties: the research domain was still in its infancy. A quarter of a century later, Worring remains fascinated by his research topic.

Prof Marcel Worring, professor Data Science for Business Analytics
Marcel Worring

‘Images contain so much information! We have been working on this for 25 years and we’re still far from being done!’, says Worring, now 49 years old and among other things associate professor at the UvA’s Informatics Institute.
He is also still somewhat of a pioneer: as a Full Professor of Data Science for Business Analytics at the Amsterdam Business School (ABS), he is one of the leading academics at the MBA in Big Data & Business Analytics that will start at the ABS next fall.
We describe the technical aspect of things, the possibilities and the impossibilities the way computer scientists see them. The business side of the MBA contributes a totally different kind of know-how: for what kind of business processes is this relevant and how does one apply this within an organisation? Most MBA’s will cover one of these two aspects. Our MBA connects both fields,’ Worring explains in his office at the ABS. 

More than just beautiful

Images are more than ‘just beautiful’, they contain a lot of information that can be of use to people and organisations with different backgrounds, says Worring. It is one of the reasons he is associate director of AmsterdamDatascience, an initiative of the UvA, the VU University Amsterdam, the Centrum Wiskunde & Informatica (mathematics and computer science research) and the Amsterdam University of Applied Sciences focusing on interdisciplinary IT-research in the field of data science. 

Worring studies pattern recognition within visual collections and their possible applications. He sees a myriad of possibilities. ‘Generally speaking, use of Big Data combined with a skilful analysis of the information obtained, facilitates better decision-making.’
A quick – automated – analysis of large numbers of images, for example, can be used for marketing purposes. Improving the results of a marketing campaign is easier when you know what aspects of the images used in the campaign have what effects, and how they can be changed, he explains. This type of analysis has long involved much more than images alone. 

Worring, who opted for computer science when he was a teenager because he was fascinated by the idea of being able to impose his will on computers, is now mainly focussed on multimedia analysis.
The internet and social media have made a treasure trove of information available that accompanies  images, often known as “meta-data”, the “data about the data”, such as geographical coordinates at which an image was taken, but also text that may be linked to image, as well as other data. Moreover, it is now technically feasible to analyse large quantities of objective and of less objective data. Anyone who does not use that information and those techniques is doing himself a disservice, Worring stresses: ‘Just looking at the image is no longer sufficient to find out what it conveys.’

Learning computers and solid visualisation

In order to teach software what kind of images go with a certain object, images are shown to the computers. ‘We’ll show for example images containing a pen and images that don’t contain one. If the computer sees enough images, it can learn how to recognise a pen.’
The same goes for text: the computer is being trained with words that do or do not apply to a certain concept. The more the different representations of the object or the concept vary, the more input the computer needs to be able to come to a trustworthy assessment.
Worring: ‘A pen always has more or less the same characteristics, just like a bottle of Coke or an Apple Ipad. With these kinds of objects, analysing one hundred pictures will lead to quite good results.’
An object like a chair or a concept such as democracy is more complicated, the associate professor adds. ‘That isn’t really a visual concept. We’ll look for pictures conveying a certain atmosphere, a connotation, which have been tagged “democracy” by others on the internet.’ The more images it sees, the better a computer will learn; but it will always have difficulties recognising a relatively abstract concept such as ‘democracy’ in a picture. 

Image recognition is merely one of the basics needed for the next level: marketing. ‘One doesn’t only want to know whether there are images of, say a bottle of Coke somewhere. One also wants to find out if these images are positive or negative,’ Worring illustrates. A picture on a blog of an overweight man drinking a Coke, for example, will evoke different feelings than that of a group of slim youngsters each holding a bottle of Coke.
In his research, Worring links the automatic analyses of unstructured data obtained from images, text and social media – multimedia mining – to structured data such as corporate results, economic and demographic data. Thus, he aims to better understand, model and improve business processes such as marketing and customer profiling. One core issue is how one can make a computer combine all these sources of information to automatically learn more. And visualising the data in the right way is just as important as finding, distilling and analysing it, Worring stresses. ‘The data should be visualised in a way that the marketing manager will understand. And in such a way that he or she can manipulate it so as to see what impact certain changes in the images are likely to have on marketing results.’ 


Already, many companies and authorities cannot do without Big Data. Worring predicts this trend will intensify as scientists and corporate circles discover more about Big Data and its possible applications.
The professor, who has also been teaching at the new Big Data Analytics track of the UvA’s Econometrics masters since last fall, did his research mostly based on theoretical business cases until now. Worring is looking forward to cross-pollination with participants of the MBA Big Data & Business Analytics. ‘This will be people with years of solid corporate experience. They’ll also ask questions to which we won’t have the answers yet. The MBA will help us to get a better insight into what companies really need in this field.’