Back Understanding Engagement, Polarization and Misinformation in AI-Based Recommendation Systems

Understanding Engagement, Polarization and Misinformation in AI-Based Recommendation Systems

Understanding Engagement, Polarization and Misinformation in AI-Based Recommendation Systems
To advance the understanding of the impact of ranking algorithms in our society, in particular how factors such as popularity, meaningful social interaction, and personalisation affect social issues such as engagement, polarization, and...

Ranking algorithms play a crucial role in literally all AI-Based Recommendation Systems. This project aims to advance the understanding of the impact of these algorithms in our society, in particular how factors such as popularity, meaningful social interaction, and personalisation, which are core features in the design of such algorithms, affect social issues such as engagement, polarization, and misinformation.

This project builds on the existing collaboration between the PI and the two co-PIs, who have previously identified a surprising consequence of popularity-based rankings with important implications to understanding the spread of misinformation [1]. In such rankings, the fewer the items reporting a given signal, the higher the share of the overall traffic they collectively attract. This effect was coined the few-get-richer effect, and it emerges in common settings where there are few distinct classes of items (e.g., left-leaning news sources versus right-leaning news sources), and items are ranked based on their popularity. The effect was shown analytically to emerge when people tend to click on top-ranked items and have heterogeneous preferences for the classes of items. The result was illustrated in simulation, showing how the strength of the effect changes with assumptions about the setting and human behavior, and tested experimentally in an online experiment with human participants.

This collaboration has been further extended by developing a theoretical framework to evaluate the effect of key parameters of ranking algorithms, namely popularity and personalization parameters, on measures of platform engagement, misinformation and polarization [2]. The framework shows that an increase in the weight assigned to online social interactions (e.g., likes and shares) and to personalized content may increase engagement on the social media platform, while at the same time increasing misinformation and/or polarization. By exploiting Facebook's 2018 "Meaningful Social Interactions" algorithmic ranking update [3], direct empirical support for some of the main predictions of the model is provided.

We plan to continue this research in three directions:

(1) Theoretical: extensions of the existing mathematical model will be developed and analyzed.

(2) Observational: A dataset has been released recently [4]. This dataset represents an excellent opportunity to test different hypotheses of the model with real-world data from Facebook.

(3) Controlled Experiments: we will conduct a large-scale human experiment where it will be possible to validate the theoretical model and analyze the impact of crucial ranking parameters such as meaningful social interaction in engagement and polarization.

[1] F. Germano, V. Gómez, G-L. Mens. The few-get-richer: a surprising consequence of popularity-based rankings. The Web Conference WWW' 19.

[2] F. Germano, V. Gómez, F. Sobbrio. Crowding out the truth? A simple model of misinformation, polarization and meaningful social interactions. arXiv preprint arXiv:2210.02248.

[3] W. Oremus et al. How Facebook shapes your feed. The evolution of what posts get top billing on users’ news feeds, and what gets obscured. The Washington Post 2021.

[4] S. Messing et al. Facebook Privacy-Protected Full URLs Data Set. 2020. https://doi.org/10.7910/DVN/TDOAPG.

----

Emma Fraxanet presents "Unpacking Polarization: Antagonism and Alignment in Signed Networks of Online Interaction" at the workshop Signed Relations and Structural Balance in Complex Systems: From Data to Models (May 15-17, ETH Zurich).

The project will be supported by the PhD Fellowship program at the Department of Information and Communication Technologies at UPF.

Gael Le-Mens is PI of several projects, including an European Research Council (ERC) Consolidator Grant (The Implications of Selective Information Sampling for Individual and Collective Judgments (#772268)), and supported by an ICREA academia fellowship (2023-2028), which can provide additional support to the execution of this project.