Machine Learning & Web Applications: Building a Scalable, Maintainable, and Reliable System for Query-based Algorithmic Political Debiasing and Bias Ranking;Perception Networks & Bias-Governing Actors: An Actor-Network Theory Lens in Post-Prison Employment;

Lahrime, Sammy, School of Engineering and Applied Science, University of Virginia
JACQUES, RICHARD, EN-Engineering and Society, University of Virginia
Vrugtman, Rosanne, EN-Comp Science Dept, University of Virginia

My STS research paper investigates the nature of post-prison employment within the United States and the contributing factors that govern the continued struggles and overwhelming lack of employment opportunities for ex-offenders. Further, my research observes and represents the issue of post-prison employment through the lens of the STS theoretical framework of Actor-Network-Theory, describing the respective social actors of employers, legislators, and consumers, as governing bodies that perpetuate a negative bias around the competency, potential, and inherent morality around ex-offenders. This negative bias creates a vicious, cascading cycle of employer hesitancy, decreased post-incarcerated outcomes, increased crime, then ultimately, increased recidivism. My STS Technical topic further elaborates on the idea of the bias-driven networks of social groups by introducing the problem frame of the increased prevalence of group polarization within online communities, specifically in the domain of search engine results and social media platforms. Group polarization is a psychological phenomenon, where individuals of a specific, shared viewpoint will gradually become more extreme with time from inner-group interaction (Fussell, 2020). For instance, individuals of a community centered around anti-vaccination advocacy efforts will likely adopt more extreme attitudes over time. Further, this has become an especially urgent issue within major social media platforms and search engines as efforts are aggressively being expanded to objectively identify biased information on a large scale. The highest risks lie in the consequences of unchecked bias, because although biased content isn’t necessarily misinformation in itself, extreme cases of unchecked bias lead to clickbait, hyperpartisan content, and pseudoscience on platforms (Timberg, 2021). For instance, in October 2022, Facebook whistleblowers brought forth a report that the platform knowingly allowed false news to be promoted to users. Further, internal evidence suggests that Facebook knew of the increased prevalence of conspiracy theories, political polarization, and incitements of violence on it’s platform weeks after the 2020 presidential election, yet took no measures to de-escalate the extreme forms of bias (Timberg, 2021). Therefore, the novel, perpetual risk of extreme bias within polarized social groups poses a major Information Retrieval challenge to platforms and search engines, that being, finding a reliable and large-scale technique to objectively quantify and identify biased information on platforms. Although up-to-date, there is no widely accepted approach to algorithmic bias-identification, there are state-of-the-art techniques in Machine Learning and content-based Natural Language Processing algorithms that show promise throughout the field. Consequently, my STS Technical Report creates a proof-of-concept solution of a web application system to algorithmically quantify the bias of a provided search query, then produce objectively ranked unbiased alternatives with the same search engine. Further, this system is designed using leading concepts in large-scale software systems design, including, scalability, maintainability, and reliability. Finally, the system was designed using a loosely-coupled microservice-based architecture, allowing for an overall increase in system reliability and maintainability, while also allowing for the implementation of different debiasing algorithmic techniques with a black-box-like approach.

BS (Bachelor of Science)

School of Engineering and Applied Science
Bachelor of Science in Computer Science
Technical Advisor: Rosanne Vrugtman
STS Advisor: Richard Jacques

All rights reserved (no additional license for public reuse)
Issued Date: