Neynar

Celebrating the experimental launch of hubs.neynar.com

Since there's been a lot of curiosity about how three different scoring / labeling models compare, I pulled some data. Here's a contour plot of the Neynar Score, vs log of OpenRank score, color coded by Warpcast Label.

Building /neynar 🪐. Former CTO of $XMTR. Mom.

Here's a scatterplot version. It's harder to see the overlap of the three categories because there are so many points covering up other points, but there's no smoothing or interpolation. This makes the spike in Neynar score at 0.5 more apparent. 

I think it looks like a whale holding a ski pole, lol.

The spike in Neynar score at 0.5 is because that is the default score for new accounts. Until there is sufficient activity in the account, the score stays at 0.5. The rationale is that, when an account has done very very little, we basically have no signal. These accounts also tend to have the lowest engagement scores.

The main conclusion is that all three models are very different. The Neynar score and Warpcast labels correlate the most, especially on users with Neynar score above 0.8 (both models agree these are not spam).

These are the correlations:
warpcast_label vs openrank_score: 0.039 (p=0.000)
warpcast_label vs neynar_score: 0.404 (p=0.000)
openrank_score vs neynar_score: 0.059 (p=0.000)

Without the log of the OpenRanks score, most of the plot is blank. This score spans a huge range and has clusters at different orders of magnitude.

Looks great!, very curious tho, how does this look if instead of taking the log you just normalised both axes? Also wonder what the correlation/mutual information is between both

Building /recaster, a third-party Farcaster client

please help check https://warpcast.com/kyletorpey

his neynar score is low, but not spammy

It's def a strong indication that an account is spammy if the Neynar score is very low AND the Warpcast label is spam. The lower the Neynar score, the higher the confidence it is spam.

Thanks for sharing this! 

If two of the models flag the user as spam, can we pretty much confirm that they are?

I like running, poker and oysters | Founder, Social Forensics | Berkeley & UPenn | Learning to listen

day job pays the bills, night job builds the future

Purple Rain

Legendary Builder | Artist | Collector | Author | Full Stack Developer • Working on /genyleap • /geny ecosystem.

Genyleap

Pi-chi.eth artist of /animeoutcasts
Hosting: /kyoto /anime-manga /farcasterunion

https://portrait.so/pichi

This is the data I’m looking for!

@geoffgolberg you might like this too!


@compez.eth and @mvr I know this will be interesting to you!

this is a nice visualization 👏 5000 $degen

Very cool/interesting chart 💪
100 $degen