Designing Social Media Recommendation Algorithms for Societal Good

Social media platforms are central for civic discourse yet evidence is mounting that they play a causal role in deteriorating social cohesion. User behaviour on social media platforms is governed by content recommendation algorithms that maximise engagement, leading to unintended consequences such as the promotion of outrage.The EU's newly enacted Digital Services Act mandates social media platforms to assess their systemic risks for society. However, the current challenge lies in translating abstract risks to concrete platform design changes that reduce such risks.We will bridge this gap by combining approaches from social science and computer science to incorporate the reduction of risk to civic discourse into content recommendation algorithms of social media platforms.To this end, we will employ a participation-based approach to develop novel algorithms that consider various aspects of civic discourse, such as information quality and diversity, and the civility of language.We will develop Open Source digital twins of social media platforms to enable experimentation with new algorithms independent of platform companies.To balance the reduction of risk to civic discourse and freedom of expression, we will solicit people's preferences in different scenarios such as a public health crisis and elections, and develop balanced algorithms.Lastly, we will develop a scenario-based risk assessment framework to assess algorithms and provide policy recommendations for interventions in content recommendation algorithms. The originality of DeSiRe stands out in that to date democratic values and fundamental rights played no role in the design of content recommendation algorithms.