#microformats 2024-02-02

2024-02-02 UTC
gRegorLove_, [tantek], gRegor, [Al_Abut], Game, vhbelvadi, [pfefferle], [KevinMarks], Guest6, BIC, [Jo] and [snarfed] joined the channel
#
[snarfed]
nudges [j12t], he's discussing elsewhere how to "determine the canonical url among several pages pairwise related by rel=me"
#
[snarfed]
do we have that? afaik representative h-card is related but not the same
#
[tantek]
snarfed, we do, via the original identity consolidation in XFN
#
[snarfed]
those are good, but I'm not seeing where they describe how to determine, of a set of pages that rel-me link to each other, which of those pages is the canonical/primary one
#
[tantek]
yeah it's implied by the star formation, that consuming code can detect stars and their centers
#
[tantek]
leaving that to already solved graph algorithms as it were rather than restating
#
[snarfed]
ah ok. but people often link to multiple of their profiles, back and forth, so often you don't have an exact minimal star
#
[tantek]
back and forth is a feature of XFN
#
[KevinMarks]
So pagerank for rel-me
#
[tantek]
point being, you should be able to detect the center by a relatively simple inbound links sort
#
[tantek]
no need for pagerank pfaffing about
#
[snarfed]
I'm confused. so you detect the center by the one with the most inbound links?
#
[snarfed]
seems ok, but the non-determinism is uncomfortable
#
[snarfed]
rel=canonical was the other proposal
#
[tantek]
that's literally what the star formation indicates yes
#
[tantek]
nah rel=canonical is for duplicate content
#
[tantek]
and does not solve anything differently
#
[KevinMarks]
The most bidirectional links. You can spam inbound links
#
[snarfed]
oh, two different things. if you _only ever_ link from silo profiles back to your web site, then yes, that perfect minimal star is deterministic and feels ok
#
[tantek]
the most inbound links from the collection you crawl obviously starting from one
#
[snarfed]
but if your silo profiles sometimes link to each other, and then you're using the one with the highest link count...I dunno. I get it, but it still feels uncomforable
#
[tantek]
no one but the big search engines (and some LLM trainers) care about the whole "crawl the whole web" scenario any more
#
[KevinMarks]
You have the Chris and Tara problem if people aren't careful
#
[snarfed]
would be nice to be more explicit, but silo profile links generally don't let you add metadata to links. I dunno
#
[tantek]
nah, that's by design. if you all you have are silo profiles that link to each other and that fails to find a center, congratulations, you have your answer, you have no center
#
[tantek]
it's very zen 🙂
#
[snarfed]
🤷 not my use case, I should defer to [j12t]
#
[snarfed]
I guess my one note would be that this idea ^ isn't on either of those pages, so if it's established and we like it, it might be worth writing up
#
[KevinMarks]
Interesting selection pressure. It would favour omg.lol over LinkTree and http://about.me
#
[snarfed]
esp the quantitative "count inbound links" part
#
[tantek]
[snarfed] feels like an obvious conclusion from first principles and math? or is that too much to ask for lol
[Joe_Crawford] joined the channel
#
[snarfed]
make the implicit explicit
gRegor, to2ds, barnaby and j12t joined the channel
#
j12t
You beat me to it Ryan :-)
#
j12t
My proposal for determining the canonical url among several pages pairwise related by rel=me would be: 1) use a star topology: the canonical is in bidirectional relationships with all others, and the non-canonical only with the canonical 2) when you come across a page with rel=me: if it has one rel=me, follow it. If it has more than one, you have the canonical one. Max one hop. Fails if you have exactly two pages in the set, but I could live
#
j12t
with that. (Or repeat the statement on the canonical one)
#
[KevinMarks]
Might be worth codifying the implicit rel=me if a silo only allows one link that was in the initial proposal
#
[tantek]
[KevinMarks] the original XFN identity consolidation page does that already
#
[tantek]
j12t also fails when people only have silo profiles that all link to each other, which is ok. like I said, in that case they have no canonical identity, the silos own them
#
[tantek]
also prototype proposals are fine, but I'd be hesitant to document any such brainstorms as "this is the way" or as a precise algorithm until folks have actually built prototypes and played with them across multiple examples to see how they work
#
[tantek]
prematurely artificially precise algorithms etc. written down a priori are nearly guaranteed to be wrong / fail when they encounter real data
j12t joined the channel
#
j12t
I'm not attempting to standardize anything, just wonder whether there is a best practice.
#
[tantek]
Best practice(s) for whom or implementing which use-case?
barnaby joined the channel
#
[snarfed]
sounded like the use case was determining the "canonical" or "primary" or "preferred" URL in a rel-me cluster
angelo_ and j12t joined the channel
#
j12t
Best practice for anybody for any use case. Just trying to understand the state of the art.
#
[tantek]
helps to document the use-cases first. [snarfed] that's close but sounds like a developer use-case, not a user use-case.
#
[tantek]
like for what user feature purpose does it matter to determine 'the "canonical" or "primary" or "preferred" URL in a rel-me cluster' ?
#
[tantek]
since users don't typically care about such cjargon
#
[tantek]
jargon*
#
[tantek]
j12t "any use case" is too broad