Politics increasingly plays out in short videos and their comment threads, but researchers who want to study who's for or against a candidate have had a data problem: most datasets keep either the video or the conversation, rarely both, and rarely the reply structure. TikStance is built to hold all of it. It gathers 161 TikTok videos and nearly 14,000 comments centered on three figures from the 2024 U.S. election cycle: Trump, Biden, and Harris.

What makes it useful is the structure. Each video links to its metadata and a threaded comment tree, so you can read stance both from the clip and from where a comment sits in the conversation. Three annotators labeled each item as Favor, Against, or None, with disagreements re-checked, and the agreement scores land in a respectable range. Notably, nested replies make up about a quarter of the comments, so the conversational depth isn't trivial.

This description comes from the abstract, so the paper is where you'll find the collection method and annotation guidelines in full.