Head of AI, AllSides · PhD, UC Riverside
I study how people argue—online and in politics. My PhD at UC Riverside focused on building NLP tools that detect arguments in text, identify what they're about, and classify where they stand. I've applied these methods to U.S. Congressional hearings, Reddit communities, and Telegram propaganda channels.
I now lead AI at AllSides, where I work on helping people encounter perspectives beyond their own.
I build NLP systems that detect arguments in text, classify their topics, and determine their stances. The WIBA framework achieves 79–86% F1 on argument detection and has been applied to congressional hearings, Reddit, and social media. ArguBias examines how cognitive bias structures cause embedding models to misjudge argument similarity, and ArguSense extends this line of work with methods for measuring argument diversity and structure in online forums.
I study how arguments differ between formal political deliberation and public online discussion. DALiSM and WIBACong compare argumentation in U.S. Congressional committee hearings (2005–2023) with Reddit threads, tracking how deliberation intensity evolves differently across these settings and revealing gaps between legislative and public reasoning.
With Mark Perry, I mapped how Russian state media uses Telegram to manufacture apparent objectivity in war reporting. We analyzed 130,000+ Telegram messages and 750 news articles, identifying distinct information laundering pathways—covert, semi-covert, and overt channels—that forward content to create an illusion of independent sourcing.
Full list on Google Scholar.
Introduces a framework and 8,000-pair corpus for identifying cognitive bias structures that cause embedding models to misjudge argument similarity. Benchmarks 10 models and shows fine-tuning reduces bias vulnerability by up to 11.6pp while improving scores on BWS and AFS benchmarks.
Introduces DALiSM, a framework for argument-centric analysis of congressional hearings and Reddit. Tracks how deliberation intensity evolves over time across formal and informal settings.
Applies argument-centric analysis to online forums with methods for detecting argument topics, describing argument structure, and measuring content diversity through clustering.
A three-part argument mining pipeline: detects whether text contains an argument, identifies its topic, and classifies its stance. Achieves 79–86% F1 using fine-tuned and prompted LLMs.
Extends WIBA to analyze U.S. Congressional hearings from 2005–2023, comparing legislative argumentation with public discourse on Reddit at the argument level.
Maps how Russian state media launders narratives through Telegram channels, analyzing 130,000+ messages and 750 news articles to identify covert, semi-covert, and overt information pathways.
An unsupervised method for expanding domain-specific keyword sets using dual embedding spaces (word-word and post-post). Evaluated on security forums spanning five years, achieving 0.82+ MAP.
Introduces Forumlyze, a framework for analyzing user beliefs in online discourse, applied to GMO discussions on Reddit farming communities. Finds Climate Change, Monsanto, and Soil Science as dominant concepts.
Tracks how sentiment and stance shift together in response to real-world events. Analyzes 7.5 million posts across 4chan, Reddit, and Parler during the 2020 U.S. election.
Open-source research tools.
Argument mining framework that detects whether text contains an argument, identifies the topic, and classifies the stance. Built on fine-tuned LLMs. Processes data from Reddit, congressional hearings, and social media. Used by 5+ research teams.
Web platform for analyzing and visualizing discourse dynamics. Upload CSV data or retrieve via API. Covers Congressional hearings, Reddit threads, and other text sources. Used at 5+ universities.