Hi! I’m a fourth-year PhD student at the University of Michigan School of Information (UMSI), advised by David Jurgens and part of the Blablablab. I work on NLP and computational social science. Before Michigan, I received a BA in Statistics and Data Science from UC Berkeley in 2023. In summer 2026 I was a research and development intern at Kitware, where I worked on personalized LLM steering.
Looking for a research internship for Summer 2027.
My research is on annotator disagreement and pluralistic judgment. Standard practice aggregates many people’s labels into a single “gold standard” and treats everything else as error. However, on subjective tasks, that disagreement is often meaningful, rooted in who the annotator is, how the task is defined, and what the item itself is. I build models that represent diverse human judgments, situated in the annotator’s identities, task, and data factors.
Selected Publications
- Xu, Y., Derricks, V., Earl, A., & Jurgens, D. (2026). Modeling annotator disagreement with demographic-aware experts and synthetic perspectives. ACL 2026 (oral). [paper]
- Xu, Y., & Jurgens, D. (2026). Beyond consensus: Perspectivist modeling and evaluation of annotator disagreement in NLP. EMNLP 2026. [arXiv]
Currently
I’m building B-HAP, a hierarchical Bayesian framework that generates per-annotator LoRA adapters from task, data, and annotator features, pooling behavior across population, subgroup, and individual levels. I’m also working on personalized LLM steering: inferring a user’s latent value profile from a handful of survey answers, then turning it into a steering vector, with active learning to choose which questions are worth asking.