Andrew Conner

I want to make a map of all the wellness clusters. There’s western science people. Then, we have the grounding/red light/nasal breathing people. But these are different from the traditional woo people (crystals, etc). Peaters, keto, anti-BCAA. Who else?

Supplement people, nootropics, peptides. Ayurveda, TCM.

I want to make a map of all the wellness clusters. There’s western science people. Then, we have the grounding/red light/nasal breathing people. But these are different from the traditional woo people (crystals, etc). Peaters, keto, anti-BCAA. Who else?

Conclusion from playing with potential What's Hot algorithms: people say they want to see cool smaller communities, but sampling randomly from smaller network clusters ends up being random and not useful.i.e., the useful pockets you'd like eventually get reflected in the follows/likes/replies graph.

What typically happens: you find a couple people in each community that are interesting (from the platform, or from other sources). Those people like/repost in their community actively, thus your second-degree graph starts to include more of the communities you care about.

Conclusion from playing with potential What's Hot algorithms: people say they want to see cool smaller communities, but sampling randomly from smaller network clusters ends up being random and not useful.i.e., the useful pockets you'd like eventually get reflected in the follows/likes/replies graph.

Been thinking about timeline algorithm design. Some thoughts, would love feedback. Algorithms are useful/needed bc the amount of content is too much even if you consider just your follows. ie, do you want every reply? every liked post? Content from people you don't follow can be good as well!

Example: there's 1000 potential posts to show you, and algo wants to show 100. It ranks the 1000 (based on distance to poster, engagement, etc), picks the top 100, and displays those in a feed. Then, importantly, they're *there* statically —

Been thinking about timeline algorithm design. Some thoughts, would love feedback. Algorithms are useful/needed bc the amount of content is too much even if you consider just your follows. ie, do you want every reply? every liked post? Content from people you don't follow can be good as well!

Social networks do this because they *want* infinite feeds that intermix new content with old. It provides an intermittent reward, even if you've read most of the existing posts. I've designed a lot of my content consumption around the concept of separating curation from consumption

Been thinking about timeline algorithm design. Some thoughts, would love feedback. Algorithms are useful/needed bc the amount of content is too much even if you consider just your follows. ie, do you want every reply? every liked post? Content from people you don't follow can be good as well!

Algorithms do two things: filter and rank. Ranking can, in effect, filter by lowering posts down a feed, but practically, low-ranked posts are never shown. Digg was the first time I remember seeing dynamic algorithmic ranking, though I'm sure it existed earlier. Reddit did it as well...

Updated my Bsky profile/follower sqlite database. Makes it pretty easy to see overall growth and what's happening in the network. Will post about follows cluster analysis later, but for now: how many unique people follow someone, per day? (Database is available if you want a copy, see below)

Would anyone like a sqlite database of the Bsky accounts and follow graph for data exploration? Includes account details (did, handle, followers, following), follows, and PageRank calculations for each account. Since no DMs here, DM me on Twitter: https://twitter.com/connerdelights

- For PageRank, only considered accounts with more than 40 followers (there's ~30 follow spammers that follow everyone). This leaves 7k accounts with a score. - I don't reflect unfollows, so if I ever got a follow between two accounts, it's still there

Would anyone like a sqlite database of the Bsky accounts and follow graph for data exploration? Includes account details (did, handle, followers, following), follows, and PageRank calculations for each account. Since no DMs here, DM me on Twitter: https://twitter.com/connerdelights

Bsky network analysis Here's how many followers the entire Bluesky network has. Nearly every account gets 20-30 followers (follow spam), and then it drops off quickly. If you have 100 followers, you're in the top 93%. The top 1% has 430 followers.

A peek into what I'm working on: a personalized, configurable trust score on Bluesky. Consider *just* people you follow and people they follow, build a PageRank trust graph. This is much smaller, and shows what accounts that your personal network trusts the most. Here's my most trusted accounts.

Bsky network analysis Here's how many followers the entire Bluesky network has. Nearly every account gets 20-30 followers (follow spam), and then it drops off quickly. If you have 100 followers, you're in the top 93%. The top 1% has 430 followers.

In fact, this seems like a decent way for future 3rd party clients to provide opt-in filters. Accounts that are significantly lower PageRank for their follower count likely have low quality followers (purchased, spammed, etc). Alternatively, this will help you find high value accounts to follow.

Bsky network analysis Here's how many followers the entire Bluesky network has. Nearly every account gets 20-30 followers (follow spam), and then it drops off quickly. If you have 100 followers, you're in the top 93%. The top 1% has 430 followers.

More PageRank analysis! Considering only accounts with more than 40 followers, here is the distribution of followers vs PageRank. This gives "trust" recursively: if your followers are largely empty accounts → low score, if largely active → high score. Explanation about what this means...

Bsky network analysis Here's how many followers the entire Bluesky network has. Nearly every account gets 20-30 followers (follow spam), and then it drops off quickly. If you have 100 followers, you're in the top 93%. The top 1% has 430 followers.

Ran PageRank on the bsky follower graph. Here are the top accounts. (My crawl is a little incomplete, but this should be directionally accurate)

Recrawling. My previous crawl method was starting from my network (and then going off without any specific limit), but I've found several follows clusters that were under-ranked just because it didn't stumble into it.

(Continuing PageRank of Bsky people, click below for context) Playing around with "who is the most underrated", I kind of like sorting by (followers rank - PageRank rank). e.g., you're the 100th most followed, and 50th highest PR. This captures how far off your follower count is from PageRank.

Similar: top PageRank people, but who have less than 200 followers (as of yesterday)

Ran PageRank on the bsky follower graph. Here are the top accounts. (My crawl is a little incomplete, but this should be directionally accurate)

Another disclaimer: to speed up computation, I only consider accounts with 20 followers. Which means if you have a lot of followers that have less than 20 followers, you don't get credit for them. Spot checking, this doesn't change things too much (and makes the graph *way* smaller). But...

Ran PageRank on the bsky follower graph. Here are the top accounts. (My crawl is a little incomplete, but this should be directionally accurate)