Revising Context, Shifting Simulated Stance: Auditing LLM-Based Stance Simulation in Online Discussions

DOI10.48550/arXiv.2606.06443MaRDI QIDQ7266556arXivFDO

Authors Jiebo Luo, Xinnong Zhang, Wanting Shan, Hanjia Lyu, Zhongyu Wei

Publication date 4 June 2026

Published in arXiv

Tldr Introduces counterfactual context revision as a framework to audit whether LLM-based stance simulations reflect genuine user beliefs. By inferring a user's stance and then revising semantically independent context, it reveals how sensitive simulated stances are to such changes, questioning their reliability.

Full work available at URL https://doi.org/10.48550/arXiv.2606.06443

arXiv classification

cs.CLcs.MMcs.SI

Cites work

• UMAP: Uniform Manifold Approximation and Projection

This page was built for publication: Revising Context, Shifting Simulated Stance: Auditing LLM-Based Stance Simulation in Online Discussions

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7266556)