~20% of overall follows are created via starter packs. arxiv.org/abs/2501.11605 by @leobalduf.bsky.social et al. - ThreadSky

suboptimal.technology • 49 days ago

~20% of overall follows are created via starter packs.

https://arxiv.org/abs/2501.11605 by @leobalduf.bsky.social et al.

Comments

suboptimal.technology•49 days ago

@leobalduf.bsky.social it might be fun to run community detection (e.g., Leiden, but it would require a chonky machine) and see if there are communities that stand out w.r.t. percentage of starter pack-created follow edges.

leobalduf.bsky.social•49 days ago

Good idea! But yeah, the complete graph is unfortunately quite large 😅 Even for the basic graph stats we have in that paper... I ran then on half a TB of memory and it took ages.

leobalduf.bsky.social•49 days ago

Also: Some people really like to mess with records! There's duplicate edges, self-loops, ... all the good stuff in there 🙃

suboptimal.technology•49 days ago

I know, I've been running Leiden in neo4j a while ago, and had to get through all that to pre-calculate relationship weights 🙂

brianell.in•49 days ago

i'm curious how end users of the "follow all" feature retain over time. twitter had something like it based on lists in onboarding, but it was net negative because people churned after ending up with noisy/confusing timeline filled with people they didn't have explicit intent to follow.

suboptimal.technology•49 days ago

Yeah, that would be interesting too. Although in the paper they only found 60k accounts *created* via a starter pack, which is probably not enough for a conclusive answer.

leobalduf.bsky.social•48 days ago

This comes with a grain of salt, though: IIRC this is a flag in the profile record, and I'm not sure it was there from the very beginning of starter packs. So this is a lower bound, for sure.

suboptimal.technology•49 days ago

As for existing accounts, they don't present activity growth for subscribers (only for starter pack creators and members).

@leobalduf.bsky.social is that because of compute resource cost or it just wasn't in your focus of interest?

leobalduf.bsky.social•48 days ago

Thinking about it some more, my gut feeling tells me that subscribers are probably more active than non-subscribers, simply because they engage so much with the network to know about starter packs...
Temporally, their engagement might also increase simply because they see "more" after subscribing.

leobalduf.bsky.social•48 days ago

Huh, no, we just didn't consider that, really. But we found two related things:

1) (so far) most (90+%) of the follow edges created via starter packs were left standing at the end of 2024. I do see the noisy timeline angle, though. I've also removed spurious SP-created follows myself before

leobalduf.bsky.social•48 days ago

2) The people "using" starter packs (i.e., pressing "Follow All") are also usually people who are members of starter packs. Anyway, as such, a good chunk of the SP users is already included in the SP-member group.

brianell.in•46 days ago

interesting! it sounds like starter packs are more of a power-user feature than i would have guessed, and are perhaps not as widely used by new users as the name suggests.

"left standing" edges is a tricky metric, as those who have a bad experience would likely leave/churn instead of manually prune

thisismissem.social•48 days ago

And sadly there's still no easy way to discover within bluesky which starter packs you are on, nor a way to remove yourself without blocking the creator

Comments

Posting Rules

Reply