LLMs for Inductive Coding
Designing ensemble pipelines where multiple LLMs propose open codes for qualitative data and a moderator model refines them into compact, human-interpretable label sets.
Computational Social Science · NLP
Angelina “Lika” Parfenova
ML researcher remixing qualitative methods with LLMs
I prototype playful workflows where language models act as collaborators, critics, and note-taking buddies for social scientists. I like messy transcripts, political debates, and making AI explain itself with codes and graphs humans can trust.
PhD @ TUM Computational Social Science
ML Researcher @ HSLU Informatik
AI Tinkerers Zurich organizer
LLMs · Qual coding · Concept graphs
STATUS
- Resting from NeurIPS
Research Clearance
AP-CSS-01
A
ID: CSS-NL-2025
Research Themes
How I use LLMs to make qualitative social science a bit less painful and a bit more fun.
Multi-Agent LLM Discussions
Simulating groups of models that discuss and negotiate codes for text, to study patterns of agreement, influence, bias, and “social” behaviour in artificial collectives.
Concept Graphs of Political Debates
Building hierarchical concept graphs from parliamentary speech, connecting utterances, codes, and higher-order themes to explore political cleavages and narratives.
Projects & Tools
A few experiments that keep my browser tabs open at night.
Pinned GitHub repo for visualizing topic models as interactive graphs—clean way to explore themes, connections, and dominant terms without living inside notebooks.
Pinned GitHub repo for brain-signal decoding experiments: model baselines, data prep, and exploratory notebooks aimed at mapping neural signals to interpretable representations.
Publications
Synced from my Google Scholar page so I don’t have to keep copy-pasting titles.
Loading latest publications...
Contact
For collaborations, talks, or sending me cursed plots.
Primary channel
angelina.parfenova@hslu.ch
Elsewhere
LinkedIn · Scholar · GitHub (links above)