IndieWebCamp SF 2015: December 3, 2015!


original post discovery is a discovery algorithm for starting with a POSSE copy of a post and finding the original post.


Use Cases

Reply to original

First, as part of How to make a comment, it would be more indieweb-friendly if post authoring implementations:

  • automatically detected when a user is trying to reply to a POSSE'd copy (e.g. a tweet),
  • auto-discovered the original post, and
  • linked to the original post instead

In question form:

  • How do I find an original post of a POSSE'd copy that I'm replying to?

Thread original posts and POSSE copies

Second, when POSSEing reply posts, it's useful to automatically:

  • markup your reply post with in-reply-to markup to the original post
  • when POSSEing your reply post to Twitter, set the in-reply-to-status-id to the status-id of the POSSE'd tweet copy of the original post.
  • for more details see: How to POSSE a comment


How to discover an original post URL from a copy of that post at a POSSE permalink

  1. retrieve the POSSE permalink
  2. in the h-entry that represents the POSSE copy, look for a link with "u-url" and "u-uid" - use that href as the original post URL.
  3. otherwise look for a rel=canonical link in the POSSE'd copy that links back to an original - use that as the original post URL.
  4. otherwise look for a link with link text of "See Original" in the POSSE'd copy page that links back to an original - use that as the original post URL.
  5. otherwise if a parenthetical permashortlink citation is the last thing in the POSSE'd copy content, convert that to a URL, use that as the candidate URL
  6. otherwise if a URL is the last thing in the POSSE'd copy content, use that as the candidate URL
  7. retrieve the candidate URL and parse it for hyperlinks
  8. iterate across hyperlinks with rel=syndication or u-syndication URLs (syndication URLs)
    1. if a syndication URL matches the POSSE permalink, then the candidate URL is the original post URL.
    2. else if a syndication URL has the same domain as the POSSE permalink
      1. retrieve the syndication URL
      2. if its redirect destination matches the POSSE permalink, then the candidate URL is the original post URL. (implementations may check such URL's redirect destinations one at a time and should stop when they find a match in order to minimize HTTP requests)
    3. end if
  9. end iteration

A parenthetical permashortlink citation looks something like:

  • ( t4Pc2)

The specific format of a parenthetical permashortlink citation is:

  1. literal '('
  2. domain name, likely short domain name (to avoid having Twitter auto-link it, as Twitter auto-links .com .net .org TLDs.
  3. literal space ' '
  4. id consisting of a-zA-Z0-9
  5. literal ')'

Convert a parenthetical permashortlink citation to URL by:

  • start with string "http://"
  • append the domain name from (2) above to the string
  • append a literal slash '/' to the string
  • append the id from (4) above to the string
  • the resulting string is a permashorturl


Algorithm Notes

Use-cases that were used to add steps to the algorithm

  • "syndication URL has the same domain as the POSSE permalink" and substeps. From the time of posting of the original post (and its POSSE permalink) to when this algorithm is run on the POSSE permalink, it's possible that the POSSE destination has changed its permalinks in some way. The following two have been seen in practice and thus are handled by this step in the algorithm
    1. http/https differences. E.g. Twitter permalinks used to be "http:" but are now (as of 2012+?) canonically "https:". Any implementation that saved POSSE tweet permalinks before that change would likely publish/link to "http:" URLs which require a retrieval of their redirect destination for comparison.
    2. change of path. Silos have in the past changed implementation specifics about how their permalinks work, leaving redirects behind for the original paths. Silos may also allow users to alter part of the permalink of a post, e.g. editing the slug, after publishing, and still support the old URL either by tracking all past permalinks for a post, or perhaps by only requiring non-post-slug portions of the permalink for unique retrieval.

POSSE Post Discovery

Main article: posse-post-discovery

Some prefer not to include permalinks/citations in POSSEd copies for aesthetic (Twitter's length limit) or technical (Instagram's lack of posting API) reasons. Is it possible to do original-post-discovery on a syndicated copy that contains no permalink or citation? posse-post-discovery. (Spoiler: yes, with syndication links, described above.)


Backlink interpretation

Implementing Backfeed has raised some additional subtleties that aren't handled by this algorithm. In particular, Bridgy generally only backfeeds responses to a POSSE post, not the POSSE post itself, which is effectively a duplicate of the original post.

However, many posts with backlinks aren't POSSE posts, and are worth backfeeding along with their responses. For example, this tweet from Kevin Marks:

@debcha @quinnnorton there, I fixed it: - need to use area for that ratio.

is a mention, not a POSSE, but it's to one of his domains, and it was posted right after the original post, so it's hard to determine that it's a non-POSSE.

Here are some possible heuristics for determining whether a backlink in a silo post is an original post link or not. These heuristics are all useful, but may not be enough for some use cases, e.g. Bridgy's high volume backfeed across many users with many different POSSE patterns.

Thanks to Kevin Marks for helping work this out in IRC!

user's domain

Is the candidate URL's domain in the silo user's profile, e.g. in the Website: field? (Ideally as a rel=me link, if supported by the silo.) If so, the candidate URL is more likely to be an original post URL.

This seems to be fairly consistent in practice, e.g. judging by experience with Bridgy's >2k users (as of Sept 2015).

Counterexample: this tweet from Armin Grewe. He has many web sites and domains, and didn't have in his Twitter profile. He added it happily, but I expect many users are in a similar situation.

Good morning with the larger version of the picture of Loch Drolsay and Glen Drolsay, Isle of Islay

Details in snarfed/bridgy#470.

24hr time window

Is the silo post within 24 hrs of the candidate URL published date (per h-entry dt-published)? If so, the candidate URL is more likely to be an original post URL.

Counterexample: This Google+ post from Ryan Barrett published at 2014-11-15 08:49 is a POSSE of this original post published at 2014-11-14 07:09, more than 25h beforehand.

near the end

Is the candidate URL "near" the end of the silo post, e.g. within four characters, or at the end after removing punctuation? If so, the candidate URL is more likely to be an original post URL.

Counterexample: this tweet from alohastone. The link is the original post link, but a hashtag comes after it, so it's not "near" the end.

Gebloggt: Statik Selektah – Beautiful Life (feat. Action Bronson & Joey Bada$$) #musikvideos

Details in acegiak/Semantic-Linkbacks#26.

edit distance

Compare the silo post's text and the original post's name, summary, and/or content, taking prefixes if they're meaningfully longer. (If the silo post has an ellipsis at or near the end, that's a strong hint to use a prefix.) The edit distance should be below a certain threshold, disregarding common differences like @-usernames in silo posts vs human names in original posts (e.g. this OP vs this POSSE).

Counterexample: this tweet from Ryan Barrett.

I want to let our kid learn from her mistakes, but my knee jerk reaction is always MY BABY IS HURTING! FIX IT!

The text in the tweet doesn't show up anywhere in the original post, a longer article that starts with:

Brooke used to have a painful head-butting habit. She’d be happily playing, then all of a sudden, she’d slam her head into my cheek…or the floor, or a toy, or anything else in range.

See Also