Improving transparency of social media algorithms using agent-based modelling
File(s)
Author(s)
Gausen, Anna
Type
Thesis
Abstract
Social media platforms have transformed how people communicate and seek information. Due to the scale of information on these platforms, recommendation algorithms have been devel- oped to sort through this information and curate what users see. This thesis introduces a novel approach for improving the external transparency of the recommendation algorithms on social media, using agent-based modelling. It has three main contributions. First, the development of an agent-based model of the social media platform X, that outperforms a published model. This model is then extended to simulate a single objective recommendation algorithm and used to evaluate the impact of different curation objectives. Second, an novel approach to improve the external transparency of the recommendation algorithm on X, using agent-based modelling. This extends the initial agent-based model to simulate a multi-objective recommendation al- gorithm that can be calibrated to mimic the behaviour of the real system. Third, acceleration of the model with both multi-core CPU and GPU implementations. This enables large scale simulations and more granular calibrations to improve the model performance. In summary, the proposed transparency approach can provide useful insights into how the recommendation algorithm prioritises different curation signals. It could be a useful tool for policy-makers to explore whether such priorities align with what social media platforms say they are doing and whether they align with what the public want.
Version
Open Access
Date Issued
2024-09-13
Date Awarded
2025-11-01
Copyright Statement
Attribution-NonCommercial 4.0 International Licence (CC BY-NC)
License URL
Advisor
Luk, Wayne
Sponsor
Engineering and Physical Sciences Research Council
Grant Number
EP/S023356/1
Publisher Department
Department of Computing
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)