The axes in the Polis visualisation do not represent any particular comment or variable.
Instead, the chart shows how different participants sit in relation to each other in an abstract ‘opinion landscape’.
There are many variables in the dataset (one for each comment), so to be able to map the opinion landscape on the 2D chart, the Polis algorithm synthesises the data through principal component analysis. The groups are defined through K-means clustering.
People who vote similarly to each other are grouped into 2-5 opinion groups, which are displayed in the visualisation.
Further information about Polis and the algorithm behind this visualisation can be found on the Computational Democracy website and in the paper: ‘Polis: Scaling Deliberation by Mapping High Dimensional Opinion Spaces’