Lee et al., "A semantic embedding space based on large language models for modelling human beliefs" (2025)

Byunghwee Lee, Rachith Aiyappa, Yong-Yeol Ahn, Haewoon Kwak, and Jisun An, Nature Human Behaviour 9(9), 1928–1940 (2025)
DOI | arXiv | PDF | Code and Data

@article{lee2025semantic,
    author = {Byunghwee Lee and Rachith Aiyappa and Yong-Yeol Ahn and Haewoon Kwak and Jisun An},
    title = {A semantic embedding space based on large language models for modelling human beliefs},
    journal = {Nature Human Behaviour},
    volume = {9},
    number = {9},
    pages = {1928--1940},
    doi = {10.1038/s41562-025-02228-z},
    year = {2025},
}

This paper uses the debate dataset to build the Belief embedding space, where co-occurring beliefs are pulled closely together. The resulting space captures how diverse beliefs interconnect and polarize across social issues, successfully predicts individuals’ new beliefs, and estimates Cognitive dissonance by measuring distance between existing and emerging positions.