Applied research · language · data systems
I turn complex questions into models, evidence, and tools people can use.
My work connects machine learning, statistics, and public-interest research. I like problems that call for technical depth, curiosity, and a wider perspective.
Chief Data Scientist · PDRI-DevLab, University of Pennsylvania
Explore reporting on democratic freedoms, civic mobilization, and restrictions. Event classification meets time-series monitoring.
ExploreFollow how foreign governments project influence through diplomacy, economic ties, and security relationships, as reported in domestic news.
ExploreConnect reporting on floods, drought, and other hazards with displacement, community action, and environmental policy.
ExploreMove from national trends to local patterns. Geoparsing and boundary reconciliation connect news to administrative maps.
ExploreThese tools measure collected news coverage; methods and source context accompany the data.
The wider projectModels, measurement & applied research
I design models, test statistical methods, and implement research pipelines. The common thread is making the assumptions visible, then checking where they break.
I built a PyTorch sequence layer that learns how to mix local attention, global latent context, and recurrent memory. Controlled tasks let me inspect when those mechanisms help—and when they don’t.
I compare six approaches to average treatment effect estimation in simulations where the true effect is known. The comparisons make tradeoffs in bias, uncertainty, and computation visible.
I develop an AI-assisted reconciliation layer that turns place mentions in translated news into consistent provinces and districts. Extracting a name is only the start; the harder problem is making it belong to the right geography.
I’m implementing an evidence-grounded coding framework for media bias. A story can report opposition to a government while framing it sympathetically; a single sentiment score misses that distinction.
I build the classification and sentiment pipeline behind collaborative work on U.S. soft power and the elimination of USAID. It connects changes in aid policy with how the U.S. government is portrayed in domestic news.
I help develop MLEED’s multilingual data and modeling pipeline. We bring environmental hazards together with displacement, community action, and policy responses, making their visibility in local news easier to examine.
Civic space · Preprint
With the MLP team, I develop multilingual infrastructure and classification models for tracking civic organizing, protest, and government action in domestic news. The work connects corpus construction with research on democratic change.
Foreign influence · Working paper
This collaboration turns domestic reporting on diplomacy, economic ties, and security relationships into monthly measures of selected influence activities. I contribute data infrastructure, model development, and public research tools.
#DataForUkraine · Collaboration, 2022
During the 2022 invasion, I helped build multilingual event extraction, geolocation, and data pipelines with Ukraine specialists and computational social scientists. Social-media reports became geographically grounded information on violence and displacement.
Illustrated previews above summarize the methods. Manuscript thumbnails here show actual first pages; local PDFs are short excerpts.
A little about me
I’m a Taiwan-born researcher and data scientist at Penn’s PDRI-DevLab. I work where language, politics, and computation meet.
Outside work, I like dancing, painting, and graffiti art. Otherwise, I’m either making coffee or on my way to make some.
A longer introduction
# A few things that feel like me
home = "Taiwan"
curiosity = ["food", "people", "places"]
off_the_clock = [
"dancing",
"making things by hand",
"painting & graffiti art",
]
coffee_status = "brewing, or about to"
small_joys = "milk tea, coffee & matcha"
Keep in touch