Research
The math behind common ground.
Polis is built on transparent, well-understood statistics — not a black box. Here's how it works, plus the papers and datasets behind it.
How it works
From votes to opinion groups
1 The vote matrix
Every participant's agree / disagree / pass votes form a large, sparse matrix: participants as rows, statements as columns. This is the raw signal.
2 Dimensionality reduction (PCA)
Principal component analysis projects that high-dimensional matrix down to two dimensions, placing participants who vote similarly near one another.
3 Clustering (k-means)
K-means groups participants into opinion clusters. Polis then surfaces the statements that define each group — and, crucially, the ones that bridge them.
Publications
Papers & preprints
- arXiv preprint 2023
Opportunities and Risks of LLMs for Scalable Deliberation with Polis
CompDemocracy et al.
Placeholder abstract. Examines how large language models can support — and potentially distort — large-scale deliberation conducted with Polis.
- Journal placeholder 2021
Polis: Scaling Deliberation by Mapping High-Dimensional Opinion Spaces
Small, C. et al.
Placeholder abstract. Describes the dimensionality-reduction and clustering approach underlying Polis’ real-time opinion mapping.
A complete, verified publication list is being compiled.
Collaborate with us
Researchers and data scientists: we welcome collaboration on deliberation, clustering, and democratic technology. Get in touch or dive into the code.