Overview
Field: NLP Authors: Yazhi Zhang, Fuqiang Niu, Bowen Zhang Published: 2025-07-16 arXiv: 2507.12492
Key points
- Political discourse has increasingly moved to short-video platforms, but computational analysis is constrained by the scarcity of datasets that jointly preserve audiovisual information and hierarchical conversations.
- TikStance is a multimodal, context-aware dataset of 161 TikTok videos and 13,876 comments designed for stance detection in political discussions.
- Coverage spans three major political figures of the 2024 U.S. election cycle — Donald Trump, Joe Biden, and Kamala Harris — with content collected from September 2023 to January 2025.
- Each discussion unit links a host video and its metadata to a parent-linked comment tree, enabling stance analysis in both audiovisual and conversational context.
- Each item was independently labeled by three annotators using a three-class scheme (Favor, Against, None) for video-to-target and comment-to-target stance; disputed items were relabeled.
- Inter-annotator agreement (Krippendorff's alpha): 0.743 for the Trump subset, 0.723 for Biden, and 0.722 for Harris.
- Descriptive analysis shows target-dependent differences in stance distribution and conversation depth; nested replies account for 23.3% of all comments.
- With multi-target coverage, hierarchical dialogue, and reliable multi-level human annotation, TikStance supports multimodal stance detection, political communication, computational social science, and context-aware NLP research.
Original abstract (excerpt)
> Political discourse has increasingly moved to short-video platforms, yet computational analysis of such content remains constrained by the scarcity of datasets that jointly preserve audiovisual information and hierarchical conversations. Here we present TikStance, a multimodal and context-aware dataset comprising 161 videos and 13,876 comments from TikTok, designed for stance detection in political discussions...
Paper link: <https://arxiv.org/abs/2507.12492>