Polarisation and Influence in Online Social Networks
Diversity-Aware Reranking of Node2vec-based Recommendations in Social Networks
T. Mihăilă (TU Delft - Electrical Engineering, Mathematics and Computer Science)
A.L.D. Latour – Mentor (TU Delft - Electrical Engineering, Mathematics and Computer Science)
M. Khosla – Graduation committee member (TU Delft - Electrical Engineering, Mathematics and Computer Science)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
Recent elections have demonstrated how social media contributes to political polarization, leading to real world consequences. Conventional "People you may know" algorithms used for new connection recommendations rely on structural similarity. These algorithms can recommend connections between people with similar opinions. In effect, this may expose users to opinions that reinforce their initial beliefs. Several methods have been introduced to reduce polarization based on interventions in the recommendation algorithm. However, none of the methods relies on post-processing the link scores. We aim to fill this gap by studying whether a diversity-aware reranking of a node2vec-based link prediction can decrease polarisation in synthetic graphs. We compare two reranking methods based on opinion and community diversity against the baseline. We look at prediction quality and polarisation change under two different opinion dynamic models: DeGroot and BCM. Our results indicate a negligible effect on polarization. However, the form of reranking has different effects based on the opinion dynamics model used.