Broadly interested in computational social science, AI safety & evaluation, NLP for social good & applications (in public health, science...)!

Happy to chat or grab coffee at the conference! Feel free to DM me :)

Forgot the most important link! Paper here:

Lastly, I would like to thank my awesome collaborators @shravika-mittal.bsky.social, Ananya Aatreya (my first mentee!), @navreetkaur.bsky.social, and faculty mentors who taught me a lot during this project @tanumitra.bsky.social @munmun10.bsky.social!

🙌 We hope public health, platforms, & researchers build on MythTriage to scale OUD myth detection on video platforms.
To support this, we’re releasing everything:
🧠 Models: huggingface.co/SocialCompUW...
💻 Code: github.com/hayoungjungg...
📊 Data: github.com/hayoungjungg...

🤩 Lastly, we’re excited because this work shows how a decade-old, but simple idea—model cascades—scales with LLM advancements to tackle real high-stakes health issues like OUD myths.

Past work tested model cascades on standard benchmarks (e.g., SQuAD). We validate them in the wild!

Our findings offer actionable insights in the context of the ongoing opioid crisis—showing the value of MythTriage:

👩‍⚕️Public health: Inform targeted interventions & debunk myths.
🛡️Platforms: Provides a scalable auditing pipeline to flag high-risk content & improve moderation.

📊Finding #3: YouTube’s recommendation continued surfacing myth-supporting content.

➡️12.7% of recs from myth videos led to more myths initially—rising to 22% at deeper levels.

⚠️ Moderation should target these rec pathways that reinforce harmful myths.

📊 Finding #2: How you filter your search results matters! Switching from “Relevance” to “Upload Date” or “Rating” increases exposure to myths—echoing the same patterns seen in my COVID-19 misinformation audit: ojs.aaai.org/index.php/IC...

😬A few clicks can change your exposure to myths!

🫶Thanks to MythTriage, we present the first large-scale study of OUD-related myths on YouTube!

📊 Finding #1: Nearly 20% of YouTube search results support OUD myths, while 30% oppose.

😰Despite more opposing, myth-supporting content is widespread—and risks shaping how people understand treatment.

⚙️So how does MythTriage perform?
📊 Achieves 0.68-0.86 macro F1 and defers only 5-67% of the examples to the costly LLM.

In practice, MythTriage:
💸 Cuts financial costs by 98% vs experts and by 94% vs LLM labeling
⏱️ Cuts time costs by 96% vs experts & by 76% vs LLM labeling

🚀 Our solution: MythTriage
👉 Uses lightweight DeBERTa for routine cases
👉 Defers harder ones to GPT-4o (high-performing but costly)

The trick? We distilled DeBERTa on GPT-4o’s synthetic labels—achieving strong performance without massive expert-labeled data.

💡Challenge: Detecting OUD myths on video platforms at *scale* is tough–clinical expertise and labeling are essential, but it is slow and costly.

🤖LLMs show promise, but high compute & API costs—especially for long-form video—limit their practicality for large-scale detection.

🩺 To rigorously detect OUD myths in our datasets, we collaborated closely with clinical experts to:

✅Validate eight pervasive myths on OUD (see examples below!)
✅Create and refine annotation guidelines
✅Build a gold-standard dataset: 310 videos labeled across 8 myths (~2.5K expert labels).

To measure the scale and prevalence of myths on YouTube, we curated opioid and OUD search queries based on real-world search interests. Using these queries, we built two datasets on YouTube:

1️⃣ OUD Search Dataset: 2.9K search results
2️⃣ OUD Recs Dataset: 343K video recommendations

🛜Facing offline stigma, many turn to online platforms (YouTube) for health info & recovery.

‼️But myths fuel treatment hesitancy, distrust in healthcare, & stigma.

🤔Understanding the scale of myths is crucial for health officials & platforms to design effective interventions.

I would also love to be added!!

Thank you for the shoutout, Joey! :)