Computational Social Science · Science of Science

Jihyang (Jiji) Cheon

I study how knowledge and expertise differ across communities — and how those differences shape collaboration, evaluation, and innovation.

About

I am a master's researcher at KAIST working at the intersection of computational social science, science of science, and sociology of knowledge. I use large-scale scholarly and behavioral data, NLP, and network methods to study how ideas, methods, and expertise are produced, evaluated, and connected across fields.

My current research focuses particularly on differences across academic fields. Disciplines often ask different questions, use different methods, and value different forms of evidence. I am interested in making those differences visible at scale and understanding what they imply for interdisciplinary collaboration and scientific innovation.

Methodologically, I work with large-scale text and network data using NLP, knowledge graphs, network analysis, and computational experiments.

My path into this work began with more applied projects on AI and society, accessibility, and generative AI perception. Those projects were less about disciplinary knowledge specifically, but they shaped a lasting interest in how technology, knowledge, and institutions interact — which now grounds my focus on how expertise itself is organized and evaluated across fields.

Project & Papers

Research Agenda

Academic fields do more than study different topics. They often ask different questions, use different methods, rely on different forms of evidence, and develop different standards for what counts as valuable knowledge.

My research examines these differences computationally and asks what happens when distinct forms of expertise meet: when researchers collaborate across disciplines, when communities evaluate the same knowledge differently, and when emerging technologies change how intellectual work is organized.

More broadly, I am interested in how the structure of knowledge shapes collaboration, innovation, and technological change.

Knowledge Across Fields

How do disciplines differ in the topics, methods, evidence, and forms of significance they value?

Collaboration Across Boundaries

When does intellectual diversity facilitate interdisciplinary collaboration, and when does it create barriers?

AI, Work, and Knowledge

How do emerging AI systems reshape the delegation, production, and organization of knowledge work?

Selected Research

Working title Ongoing research

How Do Academic Fields Define Valuable Knowledge Differently?

Do academic disciplines differ systematically in how they frame the significance and contribution of research?

Academic fields develop different intellectual norms about what makes research important. This project uses large-scale scholarly text to examine discipline-specific patterns in how researchers describe novelty, significance, contribution, and broader relevance — and whether these differences create barriers or opportunities for interdisciplinary collaboration and evaluation.

Scholarly text · NLP · Embeddings · Large-scale text analysis · Cross-disciplinary comparison · Experimental follow-up studies

More detail

Why it matters

Disciplinary norms about what counts as important research are usually implicit, carried through training and community practice rather than stated outright. Making these norms visible at scale helps clarify where interdisciplinary collaboration and evaluation succeed or break down — for instance, when a contribution framed as significant in one field reads as incremental, or illegible, in another.

Data

Large-scale text drawn from published scholarly writing across multiple disciplines. This is an ongoing project; corpus construction and field coverage are still being finalized.

Approach

The analysis combines NLP and embedding methods to characterize how researchers across fields describe novelty, significance, and contribution, then compares these patterns systematically across disciplines. A planned experimental component will test whether disciplinary framing itself — independent of content — shapes how scholars evaluate research originating inside versus outside their own field.

Findings

Findings are still being developed and will be added as the analysis progresses.

My contribution

Lead researcher, responsible for conceptual development, computational analysis, and the design of the planned follow-up experiments.

Outputs

No public outputs yet — this project is in progress.

arXiv, 2026

Who Delegates to AI? Evidence from 53,000 Agent Configurations

H. Lee, Jihyang (Jiji) Cheon, and Lanu Kim

When workers have access to agentic AI, which tasks do they actually choose to delegate?

Most research on AI and work measures either what AI could theoretically perform or what workers happen to use AI for in individual interactions. This project introduces another layer, delegated exposure, capturing tasks that users intentionally commit to reusable agentic workflows. The study analyzes approximately 53,000 user-created agent configurations from the Manus ecosystem and maps them to occupational tasks from O*NET.

53K agent configurations · O*NET occupational task data · Semantic similarity · Occupational exposure measurement · Comparative analysis of existing AI exposure indices

More detail

Why it matters

Existing measures of AI exposure largely describe technical capability or observed one-off use. Neither captures a form of adoption that may matter most for the organization of work: which tasks people choose to hand off on a recurring basis, and to what kind of system. Distinguishing capability, availability, observed use, and delegated exposure gives a more precise account of how agentic AI is being incorporated into work.

Data

Approximately 53,000 user-created agent configurations from the Manus ecosystem, mapped to occupational task statements from the O*NET database.

Approach

We use semantic similarity methods to match agent configurations to O*NET task statements, construct an occupation-level measure of delegated exposure, and compare it against existing AI exposure indices built from capability assessments or observed-usage logs.

Findings

Key results are summarized in the figures below; see the paper for the full analysis.

My contribution

Developed the conceptual framework distinguishing different stages of AI exposure and contributed to defining delegated exposure as a user-level measure of agentic AI adoption.

Outputs

Research project, 2025–2026

Forecasting Emerging Ceramic Technologies from Global Patent and Scientific Literature

Can changes in scientific and technological knowledge networks reveal emerging technological trajectories before they become established?

This project studies the evolution of knowledge around ceramic technologies — GaN, TiO₂, and Co-oxide — using large-scale scientific publication and patent data from 1970–2025. It constructs keyword co-occurrence networks from papers and patents to examine how concepts emerge, recombine, and move between scientific and technological domains.

KeyBERT / keyword extraction · Co-occurrence networks · Network centrality · Community detection · Temporal network analysis

More detail

Why it matters

Emerging technologies rarely appear fully formed. They are typically preceded by a period of conceptual recombination, visible in how keywords and ideas co-occur and migrate across the scientific literature and the patent record. Tracking these network-level changes offers a way to study technological trajectories as they form, rather than only after the fact.

Data

Scientific publications and patents from the United States, China, Europe, Japan, and Korea, covering GaN, TiO₂, and Co-oxide research from 1970 to 2025 — a multi-decade longitudinal dataset spanning both scientific and technological literatures.

Approach

Keywords are extracted from papers and patents using KeyBERT-based methods and assembled into co-occurrence networks. Network centrality, community detection, and temporal network analysis track how conceptual clusters form, recombine, and diffuse between scientific and technological domains over time.

Findings

Analysis is ongoing; this section will be updated with network visualizations as results are finalized.

My contribution

Contributes to network construction and temporal analysis within the Computational Sociology Lab, examining how scientific and technological knowledge structures evolve together.

Future direction: Graph-based forecasting approaches, including GNN models, for predicting emerging technology clusters.

Outputs

No public outputs yet — this project is in progress.

Research project

Public Perceptions and Social Trust Surrounding Generative AI

How do people understand and evaluate rapidly emerging generative AI technologies?

A computational social science project examining public discourse and social trust surrounding generative AI, part of a broader interest in how emerging technologies are interpreted and incorporated into society.

Large-scale text analysis · NLP · Computational social science

More detail

Why it matters

How the public interprets and trusts a new technology shapes its adoption, regulation, and eventual social role. Generative AI's rapid diffusion makes it a useful case for studying these dynamics empirically, rather than assuming a single, uniform public reaction.

Approach

The analysis applies NLP and computational social science methods to examine how generative AI is discussed and evaluated in public discourse.

Findings

This project is ongoing.

My contribution

Conducted as part of research at the Social Data Science Lab, POSTECH, examining public perception of generative AI technologies.

Research project

Understanding Barriers in Sign-Language Interpretation Services Through Complaint Data

What institutional and interactional barriers emerge when sign-language interpretation services respond to high-stakes or emergency situations?

Analyzed complaint records from sign-language interpretation counselors to identify recurring institutional and interactional barriers, including challenges related to emergency response, abusive interactions, and emotional labor — part of a broader interest in using computational methods to study inequality, institutions, and social systems.

Administrative complaint records · Text analysis · Topic modeling · Qualitative interpretation

More detail

Why it matters

Interpretation services sit at the intersection of accessibility, public infrastructure, and emotional labor. Complaint records offer a systematic window into where these services fail users in practice, particularly in time-sensitive or high-stakes situations — evidence that is otherwise difficult to observe directly.

Findings

Findings point to recurring barriers around emergency-response protocols, abusive interactions, and the emotional labor placed on interpretation counselors.

My contribution

Conducted as undergraduate research at POSTECH.

Papers

Published / Preprints

  • [1]

    Lee, H., Cheon, J., & Kim, L. (2026). Who Delegates to AI? Evidence from 53,000 Agent Configurations. arXiv:2608.20425.

    arXiv preprint
    Paper arXiv

Working Papers / Ongoing Research

  • [2]

    Cheon, J. How Do Academic Fields Define Valuable Knowledge Differently? Working paper, in progress.

    Ongoing research
  • [3]

    Cheon, J. Forecasting Emerging Ceramic Technologies from Global Patent and Scientific Literature. Research project, 2025–2026.

    Ongoing research

News

News

2026
Preprint released: “Who Delegates to AI? Evidence from 53,000 Agent Configurations” (with H. Lee and Prof. Lanu Kim).
2025
Joined the Computational Sociology Lab at KAIST as a graduate researcher, advised by Prof. Lanu Kim.
2025
Completed undergraduate research at the Social Data Science Lab, POSTECH, on accessibility and generative AI perception.
2022–2025
Co-founded Cream Co., Ltd., building a generative-AI image tool for webtoon creators; selected for the Korean TIPS program.

[More updates added as research progresses]

Education

Education & Research Experience

2025–Present

KAIST

Graduate Researcher — Computational Sociology Lab

Advisor: Prof. Lanu Kim

  • Discipline-specific patterns in scholarly significance framing
  • Interdisciplinary research
  • AI delegation and work
  • Computational social science
2024–2025

POSTECH

Undergraduate Researcher — Social Data Science Lab

Advisor: Prof. Young Bae

  • Accessibility and public services
  • Generative AI perception and social trust
2020–2022

POSTECH

Undergraduate Researcher — Design Intelligence Lab

Advisor: Prof. Juhong Park

CV

Curriculum Vitae

A web summary of my academic background. For the complete, citable version, download the PDF.

Download PDF [ADD CV PDF]

Industry Experience

2022–2025

Cream Co., Ltd.

Co-founder

  • Developed a generative-AI image creation product for webtoon creators.
  • Raised approximately KRW 500M in funding; selected for the Korean TIPS startup program.

Included as additional experience with real-world AI adoption and human–AI interaction — not a primary research line.

Research Toolkit

Text and Language

NLP · Embeddings · LLM-based analysis · Topic modeling

Networks and Knowledge

Network analysis · Knowledge graphs · Community detection · Temporal networks

Empirical Analysis

Statistical analysis · Experimental design · Survey analysis

Computational Tools

Python · R · PyTorch · Transformers · Scikit-learn · SQL

Awards & Academic Activities

  • Excellent TA, KAIST
  • KOSSDA Social Data Analysis Competition — 2nd Place
  • POSTECH Undergraduate Research Program — 3rd Place
  • Chey Institute Travel Grant
  • Korean Government Scholarship for Science and Engineering

Teaching

  • Teaching Assistant, KAIST — recognized with the Excellent TA award.

Interested in similar questions?

I am always happy to talk about science of science, interdisciplinary collaboration, computational social science, and the changing organization of knowledge work.