Projects are listed according to their readiness for submission (if they are not already under review). For collaborative work, my contribution is often (but not always) the computational analysis of the data.
I test 40+ tiny (<1B parameters) and small (>1B & <4B parameters) language models to show they outperform widely used small language models on most (15/17) general social science tasks. In preparation for Sociological Research Methods.
I created (with Scott Gartner and Diane Felmlee) an end-to-end pipeline that automates real-time global political violence data collection using open-source LLMs with humans in the loop. We demonstrated the pilot at the International Studies Association, in which we replicated and extended the Global Terrorism Database (GTD). Manuscript in prepration.
Using a dataset of roughly 50 million songs and various NLP techniques, I analyze the limits of innovation in the population by examining how audiences evaluate songs that are sonically and semantically atypical, and how producers can position their products strategically to avoid categorical illegitimacy discounts. In preparation for Organization Science.
Using NLP techniques such as embeddings and topic modeling with ~30,000 articles from top general sociology journals since 1923, my co-authors (David Baker and Enes Atac) and I find that stratification has become the main analytical frame of Sociology (deployed in as much as ~65% of all publications in 2025).
Using a dataset of roughly 5 million tweets from 2009 to 2022, my co-author (Enes Atac) and I trace how political and economic dynamics drive anti-refugee hate in Turkey. In preparation for the American Sociological Review.
Using a dataset of roughly 12 million lyrics and LLMs as data extraction tools, I find that while gender stereotypes are diluting across societies, they are paradoxically intensifying in popular music (independent of the genre effects i.e., the rise of hip-hop/rap).