Abstract

Contemporary social media is polluted by the viral diffusion of low-quality and occasionally false content. More worryingly, in the political debate this phenomenon takes the form of what is usually called fake news, which mislead public opinion. Actually, some scholars have proposed the more apt name of junk news, with a reference to “junk food” in order to emphasize their low quality, rather than their actual falsity, and their ability of distracting instead of nourishing the public debate.
Since digital platforms are man-made, researchers have sought to locate the responsibility for these information disorders in the platforms designs. Indeed, these pathologies appear to be the byproduct of the very nature of online social networks, that make interactions highly dynamical and introduce unprecedented effects of feedback, speed, and scale. The underlying motivation for platform designs that favor pathological attention dynamics could be found in the attention economy that underpins the business models of most digital platforms. As already identified by the seminal work of Goldhaber, the economics of digital platforms rely on attracting and maintaining the attention of the users, which is monetized through advertisements. To this purpose, the digital platforms record users’ behaviors and use sophisticated algorithms to select which information is most suitable to increase their engagement with the platform.

The main culprit has therefore been identified as the recommendation systems that power digital platforms. Recommendation systems are ubiquitous in web services like social networking service. Their purpose is sieving the available information and directing the user tothe most relevant content. Recommendation systems leverage a wide array of machine learning tech-niques, which allow not only to quantify the absolute relevance of the items but also to tailor the recommendations to the expected tastes of the users, whose online behaviors are suitably recorded

Building upon this reflection on the dynamics of contemporary social media and of the role of recommendation systems therein, I have recently constructed a simple but insightful mathematical model of the user-recommendation system interaction. This model features a user that interacts in closed loop with an online news aggregator, such as Apple News or Google News. My preliminary analysis of this simple model indicates that recommendations distort the opinions and this distortion is strongly correlated with the click-through rate achieved by the recommendation system (improving the click-through rate makes the distortion larger). In terms of management, the distortion effect can be mitigated by increasing the randomness in the recommendations, albeit with a trade-off with the click-through rate.

In a broader perspective, my work advocates an integrated approach of mathematical modelling, network science, and control systems theory with the purpose of understanding and managing the pathological attention dynamics of online media.

Castaldo, M., Venturini, T., Frasca, P., & Gargiulo, F. (2020). Junk news bubbles modelling the rise and fall of attention in online arenas. New Media & Society https://journals.sagepub.com/doi/abs/10.1177/1461444820978640

Rossi, W. S., Polderman, J. W., & Frasca, P. (2021). The closed loop between opinion formation and personalised recommendations. Conditionally accepted in IEEE TCNS https://arxiv.org/abs/1809.04644