https://github.com/compdemocracy/polis/discussions
- “No, make own instance”
Divisiveness/consensus metric in automatic report · compdemocracy/polis · Discussion #1661 · GitHub
- “divisiveness” = “extremity”
- Polis: Scaling Deliberation by Mapping High Dimensional Opinion Spaces p.9-10
- I’m a little unsure about this, but it’s not necessary right now, so skip it.
- The non-moderated style detracts from the experience of those who join later.
- Moderation is hard, needs improvement.
- Ability to edit and approve opinions submitted
- Need to ensure transparency in moderation.
- Dissent is usually not necessary.”
- This is something we need to discuss further.”
- It’s better to have both sides of the story.
- The theory that it is easier to understand to characterize a group as “agreeing” than to characterize it as “disagreeing”.
- gain in quantity
- Save your remarks
- It’s better to have both sides of the story.
- The problem that the opinion moderation screen was 100 new posts with no pagination, so moderation was no longer possible.
- This is something we need to discuss further.”
- Categorize statements by topic to facilitate de-duplication
- Maybe we can determine this by the similarity of the embedding vectors.
- Or, if there are 10-20 users who do not feel burdened with voting, the alternative could be to let them vote first, and then ignore them if their voting result vector is less independent of the existing opinion vector (less likely to provide new information).
- This might also relate to the existing “Priority” mechanism
- Written by.
- Maybe we can determine this by the similarity of the embedding vectors.
Why is there 109 people voted but only 49 grouped? · compdemocracy/polis · Discussion #1528 · GitHub
-
Participants have to vote a minimum of 7 times to be included.
Accessing the pol.is vizualization · compdemocracy/polis · Discussion #1282 · GitHub
- How the comments themselves are embedded in the 2-D plane
-
In short, the comments are projected as though they were a participant who voted on only the comment in question. So in effect, they’re just projected according to the PC loadings, scaled by the same factor applied towards participant projections, so that they can be more easily plotted together.
- In essence, they are projected as if they were participants who voted only for the comment in question. So, in effect, we are just projecting according to the PC loadings and scaling them by the same coefficients that are applied to the participant projections so that they can be more easily plotted together.
- mathematics
- Conversa Polis User Group open calls - HackMD
- No voting accompanies the moderator view.
- If an enthusiastic participant wants to elicit the approval of a group other than his own, he needs to write in the negative form if he knows that he will be implicitly +1
- “not not A != A”
- Trying to convert to positive wording on GPT-3 only on report page
- GPT API is personalized and needs to be optional.
- I’m doing a PCA and then calculating the distance, why? Aren’t you throwing away information?
- PCA acts as noise reduction
- I’m focusing on information useful for 2D visualization.
- But you don’t want to miss what’s happening in the third dimension and beyond.
- I took the top four dimensions of the PCA and made a 2D plat of each, but there was a surprising correlation.
- No voting accompanies the moderator view.
- Conversa Polis User Group open calls - HackMD
-
I started a version to show how groups are created and comment chosen according to the original paper, however, not all of the methodology is clear to me from
This page is auto-translated from [/nishio/Polis Discussion](https://scrapbox.io/nishio/Polis Discussion) using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I’m very happy to spread my thought to non-Japanese readers.