Zachary Levonian


Machine Learning Engineer, Data Scientist, and Researcher
PhD in Computer Science
Envelope zacharylevonian at gmail dot com

My research lies at the intersection of social computing and natural language processing. I am interested in the way humans communicate with and support one another online.

I completed my PhD with GroupLens at the University of Minnesota in October 2022. I was advised by Professors Lana Yarosh and Loren Terveen.

My current research is in the domains of education and human-AI interaction.

Quick links: LinkedIn | GitHub | Mastodon | Twitter | Bluesky | Blog | Google Scholar



Research
Useful Machine Learning

A major theme of my research is creating machine learning models that are useful to humans. That starts with developing effective annotation and evaluation practices grounded in the detailed understanding of a problem. In my experience as a data scientist, I've written about how to incorporate expert knowledge during problem formulation and annotation [ICWSM 2020], how to evaluate models when very little data is available [IUI 2022], and how to create metrics appropriate for a specific user interface [CSCW 2025].

I am also a believer in open-sourcing research code: I've released Python packages for multi-modal deep learning recommender systems, for simulation experiments with active learning, and for retrieval-augmented generation with large language models.

Social Support in Online Health Communities Haiwei Ma viewing CaringBridge on a laptop

Online social support groups are important places for patients and caregivers to seek information, express themselves, and exchange support. I collaborated with CaringBridge, a prominent online community for writing about and sharing personal health journeys, to study these online health communities. My most recent work investigated how to foster supportive relationships between CaringBridge authors via peer-to-peer recommendation [CSCW 2025]. Our work has been cited in the Star Tribune [1, 2], with publications accepted to [ICWSM 2020], [CSCW 2020], and [TOCHI 2020].

Automatic Speech Recognition for Aviation Haiwei Ma viewing CaringBridge on a laptop

I worked with a research team at The MITRE Corporation to create real-time safety monitoring systems for air traffic control towers. These systems use automatic speech recognition to detect radio misunderstandings. We presented our work at [DASC 2015] and [ATM 2017]. I also received a fellowship through MITRE's Early Career Research Program to explore natural language understanding of air traffic controller and pilot radio transmissions.

Publications

Google Scholar profile

Star (*) indicates students I mentored.

Posters & Non-archival Publications

Publications in this section received peer review for non-archival venues. See my blog for self-published writing.

Star (*) indicates students I mentored.


Blog

View all of my self-published writing here.

Most of my blog-style writings are on GitHub or Medium.

What's New
Recent research updates. I keep my professional updates on my LinkedIn.
Miscellaneous

In my spare time I visit breweries, see weird indie shows, and try to predict the outcome of The Bachelor. I have competed in exactly one curling tournament, where I was laughed at by a former Olympian. I am an occasional editor on English Wikipedia (2000 edits and counting).

I go by Zach or Zachary and my preferred pronouns are he/his/him.

I am on Mastodon, Twitter, Bluesky, LinkedIn, and GitHub.

This website is hosted on GitHub.

This page last edited on October 17, 2024.