A computational framework for auditing targeted advertising

• Autor(a) principal: Márcio Aparecido Inácio da Silva
• Data de Publicação: 2024
• Tipo de documento: Tese
• Idioma: eng
• Título da fonte: Repositório Institucional da UFMG
• Texto Completo: http://hdl.handle.net/1843/68158; https://orcid.org/0009-0000-1214-4245
• topic: Misinformation; Political ads; Social networks; Transparency mechanisms; Computação – Teses; Redes sociais on-line – Teses; Desinformação – Teses; Propaganda política – Teses

Since the 2016 United States presidential election and the Cambridge Analytica scandal, political sponsored content has become an effective form of political campaigning. However, this election was marred by the abuse of targeted advertising on Online Social Networks (OSNs). Concerned about the potential for similar abuse in the 2018 Brazilian elections, we designed and deployed a computational framework to instantiate targeted advertising audit systems to detect political ads on OSNs. Firstly, in response to the abuse of targeted advertising on Facebook during the 2016 United States Presidential elections, we selected the Facebook platform for validating our framework. To achieve this, we adapted a browser plugin to collect ads from the timelines of volunteers using Facebook. We successfully enlisted the support of over 2,000 volunteers for our project who installed our tool. Subsequently, we employed a Convolutional Neural Network (CNN) for the detection of political Facebook ads using word embeddings. To assess the effectiveness of our approach, we manually labeled a dataset of 10,000 ads as either political or non-political. We then conducted an extensive evaluation of our proposed approach for political ad identification by comparing it with classical supervised machine learning methods. In conclusion, we designed a computational framework for auditing targeted advertising and we instantiate this framework in different scenarios (\eg Facebook ads, public posts from Facebook groups, Facebook page posts, and Twitter posts). %two real-time system that displays Facebook ads identified as politically relevant. Notably, on Facebook we observed that not all political ads we detected were present in the Facebook Ad Library for political ads. Even though Twitter posts, as well as posts in groups and pages on Facebook, are not promoted, they can still be orchestrated to create an amplification effect. Our findings underscore the significance of enforcement mechanisms for declaring political ads and the imperative need for independent auditing platforms.