1. médialab Sciences Po
  2. Actualités
  3. Duncan Cassells, mobiliser les modèles de dynamique des opinions pour comprendre la polarisation

Duncan Cassells, mobiliser les modèles de dynamique des opinions pour comprendre la polarisation

Duncan Cassells, doctorant au département d'informatique (LIP6) de Sorbonne Université, associé au médialab, a soutenu sa thèse intitulée « Comprendre la polarisation : relier la dynamique des opinions aux distributions empiriques et aux groupes » en décembre 2025. Dans cet article, il revient sur son parcours de thèse.

Chronique

Can you first introduce yourself and describe your academic background?

I studied at Bristol University in the UK, I have a Masters in Engineering and Mathematics. After my Masters, I was interested in doing a PhD but I wasn't sure about the subject of study, so I took some time to think. Meanwhile, I worked professionally, first as a financial data consultant and then as a data analyst for the editorial team at the Guardian newspaper in London. During that time, I looked at places that might suit me for proposing a PhD. I wanted something that was interdisciplinary, so that I could apply my skills in mathematics and computer science in a social science context. I looked at the medialab and it seemed like an interesting place. I saw the seminars and kept up to date for a while to see what kinds of subjects were being talked about, and then I contacted Pedro Ramaciotti and Jean-Philippe Cointet. I was successful in getting funding from Sorbonne University and then associated here at the medialab.

Can you introduce your thesis topic?

The title of the thesis is “Understanding Polarisation: Bridging Opinion Dynamics with Empirical Distributions and Groups”. The main focus of it is using opinion dynamics, that is models of how individuals’ opinions change, as a tool to understand polarisation.

The first strand of work in the thesis was on including groups in opinion dynamics models - in response to a criticism of current models that they treat individuals as identical. That's saying that every person that you interact with, you can treat exactly the same, which is a big assumption. In order to address this, I included a conception of groups in the model, whereby individuals identify each other as in-group or out-group, and treat each other differently depending on that identification.

The second avenue was a collaboration with some colleagues at Sorbonne Université, who are also part of INRIA, to link towards empirical data. It’s quite easy to come up with models, but to actually test them against opinion data is quite a complicated thing, because how do you observe opinions? How do you get opinions over a long period of time with a consistent population? How do you connect models of individuals to large social media datasets? This part of my thesis looked to address a perceived lack of empirical validation in the current opinion dynamics literature. We did this by moving from what's called an agent-based model, to instead modelling a distribution of opinions, which then allows us to use large online datasets, such as from a social media platform.

What got you interested in this subject?

This study was a way to approach problems such as polarisation of opinions with the skills I had from my background in complex systems and mathematical modelling. I could then try to improve on those skills with knowledge from exposure to people from other disciplines, which is where the medialab really came in.

What methods did you use for your research?

The methods that I used were certainly more quantitative. The opinion dynamics work was simulation of models and proposing model validation frameworks. I would build a model in Python, or adapt a model that was in Fortran and use that with a Python wrapper. The final part of the thesis was connecting these simulations to empirical data from the European Polarisation Observatory.

What are the main contributions of your thesis?

Each of the two strands of work presents a different way to consider opinion dynamics models, starting with this idea of groups, which is a new method within the modelling world. It also showed that the consideration of groups within these models produces a different outcome. Outside of the modelling world, it's certainly important to acknowledge something like affective polarisation, but you need an idea of groups to be able to work on this. So the first contribution is introducing a method to deal with groups and allowing the consideration of affective polarisation within these models.

The second contribution is enabling the validation of these mathematical models against large datasets by applying a method by which you can simulate distributions of opinions to connect with large datasets. The method also maintains the ability to follow the opinion change of average profiles of individuals so the models can be evaluated at different scales. It facilitates an approach to the empirical validation of the models, which is thought to be lacking at the moment within the research field.

Did you encounter some difficulties during your research? If so, how did you overcome them?

Initially I thought beginning the PhD would be hard, coming from a professional work environment and moving to an academic work environment in a new country. Of course I had to learn and adapt, but it was not so hard because people are friendly, and it's a welcoming environment.

Then, producing a thesis is a big task, and I think there's a difficulty there in terms of how you break this down. You have to manage this process, and create milestones as you go along. I think this is really helpful, because the amount of work ahead of you can feel overwhelming.

How did you organize your work between the labs?

During the thesis, I spent time with the Complex Networks team at Sorbonne University, as well as here at the medialab, and then also at the Learning Planet Institute. I tried to spend at least one day a week at each place.

For the medialab itself, I'd try to prioritize the team lunch meetings and the seminars. The lunch meetings are a chance to meet people in a more casual situation, where some work is discussed, but you also see the life of the labs. The seminars were really interesting to be exposed to a wider variety of research approaches and ideas. They are a real idea generator in some way, it allows us to sit, think and be exposed to new things, which can create new ideas.

How would you describe your experience as a PhD student?

It was very positive overall. I think I really enjoyed having the time to develop my thinking about how research should be done, how to approach and frame a research problem, and how to plan the work after that. Having the time to do this was really great, without being focused on one goal in particular. Over the period of the PhD, there is a final goal, which is always there, but in each year or each month, you have some time to follow an interest or learn about something else, and develop your ideas. The freedom to do this was really the thing I enjoyed the most about the PhD.

Also, having a research team, an engineering team and a design team is a really nice aspect of the medialab and something that was very enjoyable to experience. It encourages openness to different approaches and you have a wide range of experiences.

Do you have any advice that you would give to someone who wants to start a PhD?

In my mind, I viewed the process as a funnel, where your focus is very wide at the start, and then it narrows as you go forward in time. I initially wanted to keep my focus wide, I enjoyed learning about lots of different things. But then narrowing was also an important lesson of focusing on something and getting your research objectives done and complete. So being open at the start, thinking, taking your time to read widely, being exposed to new concepts, and then narrowing to the final product, and then being able to open up again afterwards.

What are your plans now? Are you continuing with research?

I have an article to write from finishing my thesis, so at the moment this is one of the main focuses of my time.

I am also looking for a postdoc position. Perhaps something that would be relevant to my experience of being somewhat interdisciplinary and working with a wide range of people, which I think a postdoc and following in research gives the freedom to do.