Research

DeSiRe - Designing Social Media Recommendation Algorithms for Civil Discourse

I'm currently working on the DeSiRe project, a 5-year ERC-funded initiative led by Prof. Jana Lasser. The project tackles a challenge close to my heart: how can we transform social media from a source of societal tension into a tool for constructive dialogue?

A core aspect of social media that we are addressing is that current recommendation algorithms are designed with a single goal: maximising user engagement. While this seems innocent enough, it often leads to undesirable outcomes, like incentivising inflammatory content. Our work aims to fundamentally rethink this approach.

This project is highly relevant to important developments in Europe, namely the EU's Digital Services Act (DSA), which requires platforms to assess and mitigate their societal risks. We're working to bridge the gap between these regulatory requirements and actionable technical solutions. By combining social science insights with computer science methodology, we're developing recommendation algorithms that actively promote civil discourse while preserving freedom of expression.

Some research questions:

• What constitutes "healthy" civic discourse, and how can it be measured?
• Can alternative recommender systems aid cultural evolution online?
• Can engagement-based algorithms ever be good for society?

Belief Networks Evolution: Analyzing the Belief Structure of U.S. Americans

This project investigates the evolution of "belief networks" over time using the General Social Survey (GSS) data. In simple terms, we are learning how different beliefs "interact" with each other and evolve across the decades in the United States. This involves isolating and capturing correlations between different beliefs, and seeing how these correlations change over time, and the emergent structure they are a part of.

A particularly interesting idea originating from this work is the notion that different sub-populations (like liberals and conservatives) actually have different belief network structures. That is, the way in which these groups relate ideas are different. A large effort of this project is to interpret and measure this effect, and to see what implications this might have for broader socio-political dynamics.

At some point, I'll add some nice visualisation here.