Abstraction and application: complementary perspectives on sociotechnical systems

Bajpai, Tanvi

Permalink

https://hdl.handle.net/2142/132498 Copy

Description

Title

Abstraction and application: complementary perspectives on sociotechnical systems

Author(s)

Bajpai, Tanvi

Issue Date

2025-11-19

Director of Research (if dissertation) or Advisor (if thesis)

• Chekuri, Chandra
• Chandrasekharan, Eshwar

Doctoral Committee Chair(s)

• Chekuri, Chandra
• Chandrasekharan, Eshwar

Committee Member(s)

• Sundaram, Hari
• Venkatasubramanian, Suresh

Department of Study

Siebel School Comp & Data Sci

Discipline

Computer Science

Degree Granting Institution

University of Illinois Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• abstraction-first
• application-first
• methodological spectrum
• sociotechnical systems
• algorithmic design
• real-world contexts
• fairness in clustering
• approximation algorithms
• theoretical computer science
• human-computer interaction
• k-center
• facility location
• covering problems
• online community moderation
• platform governance
• technological affordances
• Reddit moderation
• modqueue
• human-in-the-loop systems
• sociotechnical complexity
• design interventions
• algorithmic tradeoffs
• empirical grounding
• technological abstraction
• real-world use
• user needs
• context-aware design
• feedback loops
• research methodology
• algorithmic fairness
• community governance

Abstract

Algorithmic and technological systems are often developed using abstract models and objectives, but are ultimately used for specialized real-world applications; the gap between these levels of design and use often leads to unmet user needs or amplified harms at scale. Addressing this gap benefits from examining the two ends of the methodological spectrum for adapting generalized technologies to specific real-world contexts, namely abstraction-first and application-first approaches. This dissertation examines these perspectives through two parts, each comprising three projects that share a common methodological lens and research focus. While each part explores a different problem space, they offer complementary insights into how generalized technologies can be adapted to specific real-world settings. The first part examines the abstraction-first perspective, in which algorithms are developed according to abstract models that are modified to capture potential realities of actual use. It begins with classical metric clustering problems such as k-center and facility location, which often underpin machine learning and resource allocation algorithms where real-world concerns such as fairness have become especially relevant. The first two chapters develop constant-factor approximation algorithms for clustering formulations that aim to ensure fairness for each individual point being clustered, later extending these results to address fairness at the group level as well. The final chapter broadens the scope to the more general class of covering problems, analyzing fairness-constrained formulations under assumptions motivated by real-world applications. These results yield constant-factor approximations in settings where no such bounds were previously known and, in some cases, improve existing guarantees. While this perspective contributes flexible results that can be applied across a wide variety of contexts, it does not provide guidance for practical use; rather, it demonstrates the feasibility of incorporating practical considerations into classical models without prescribing how they should be applied in real-world settings. The second part takes an application-first perspective that explores real-world contexts to identify opportunities for technological abstraction or intervention, focusing on the domain of online community moderation. The first chapter in this part examines the broad landscape of platforms and the technological features that shape their governance structures, and the following two chapters focus on a specific platform and feature: Reddit’s moderation queue, which enables community-specific human review of reported content. This perspective illustrates how technological abstractions can emerge from studying real-world practices and highlights the factors that shape the design and use of moderation tools. While it offers insights into the mechanics of sociotechnical systems and informs design and policy interventions, these insights are inherently limited in generalizability and longevity as applications and user practices continue to evolve in ways that can transcend existing abstractions. Placing these two bodies of work side by side, each grounded in a distinct methodological paradigm, underscores how each frames, constrains, and enables different kinds of progress within sociotechnical research. This dissertation concludes with broader reflections on abstraction- versus application-first perspectives and emphasizes that neither perspective is sufficient on its own; effective inquiry depends on feedback loops between them, where abstractions are continually grounded in real-world contexts and empirical insights inform new abstractions. It also outlines avenues for integrating these methodological perspectives and highlights the importance of addressing the tradeoffs inherent in building technologies and theories that are both principled and aligned with the complexities of use.

Graduation Semester

2025-12

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/132498

Copyright and License Information

Copyright 2025 Tanvi Bajpai

Owning Collections