The delights of corporate culture are on show in a post from Scoble :

Steve Gillmor’s report on Attention.xml is included in Esther Dyson’s Release 1.0. Thanks to Mike Manuel for letting us know the report is now available for $80. I’ll have to check our corporate library and see if it’s available there (I believe it is).

Then of course there was an (invitation-only) FOO Camp presentation…
Sod all that. Here’s my €0.02 report.

Right, hitting the spec: I think the problem statement is a bit light, but there certainly is an interesting problem in this area around “with so many unread items, how do you know which to read first”. This is one aspect of the challenge facing anyone working with syndicated material - how to filter and focus the info overload down to what is really of interest to the individual end user. The notion of attention has a lot of potential here.

The sample application profiles provided help suggest where the information might come from (it would be helpful to push some of this info up into the spec intro, in an overview fashion). But Attention.xml basically provides a way of describing aspects of a user’s visits to a blog/feed/page/post/item/entry in a machine-readable fashion. This is information that could be extremely useful if captured, to both clients and servers of feeds. Doesn’t really matter where it comes from.

Next, I’d suggest that they have done a pretty good job of identifying the values that would be of use in the context of “attention". However, I do believe there are systematic problems that leak between format and model. Basically the model is the format, and the both tied to a hierarchy. While trees may be useful for many things, they are a handicap on the Web where the default structure is a directed graph.

Rather amusingly the spec answers a FAQ question about the MeNow work as “premature, and hugely overdesigned (like most RDF efforts)”. It’s hard to judge the “premature” of this - it’s relatively new, who knows whether the world is ready for it… I believe the folks behind it have been simultaneously playing with implementation code, but I think it’s still pretty much untested. Overdesigned? Well both MeNow and Attention.xml appear to have come from a brainstorming session where had I been present, they would both now have an element “catPhoto".

[PS. “Like most RDF efforts” - rather an unfair generalisation, although it certainly can go this way. Because you can add as little or as much information to a schema/vocabulary/ontology as you like, it can be difficult to know it’s time to stop designing and get on and do some application implementation stuff with your creation. I’ve got the balance wrong more than once myself.]

The use of the XOXO format to me seems bizarre. Now I have great admiration for Tantek and co’s approach in trying to remove the ambiguity of OPML, and tighten up the semantics in the process. But just stand back for a moment. Assuming you’ve decided you’re going to use a hierarchical model, and design your own XML format for it, why not build one that fits the task more closely?

Look what happened elsewhere in the context of “Simple” RSS. The use of OPML for things like blogrolls appears to have been in part to Dave Winer’s almost religious obsession with Outliner apps, part due to what was easy for Userland to code in their products, part due to Dave’s promotion and the lack of any obvious alternative. As a channel list format I reckon OPML leaves a lot to be desired, but it mostly works. It works not because it’s a good format for the job, but because the environment in which it is deployed is extremely flexible. On a technical level protocols and formats like HTTP and HTML/XML allow a huge berth within in which functional components can operate, and to an often surprising extent interoperate, despite loose languages.

No doubt, XOXO is a nice way of expressing hierarchical documents. Not long before XOXO appeared, I’d personally reached a very similar conclusion on how best to do outliner markup. XHTML brings all the benefits of XML, convenient renderers (browsers), the marvels of CSS. The thing is though, what’s being expressed by Attention.xml isn’t a document aimed at direct human consumption, but a bunch of structured data destined for machine processing. Why complicate matters by wrapping it in a document format? I’d say it falls between two stools - it won’t be all that interesting/useful as a document in a browser no matter how prettily styled; it won’t be all that good as input for program control or reasoning as the language semantics aren’t well-defined. But like many other formats, it almost certainly is good enough on both counts.

The blogroll/channel list format isn’t the only place OPML has been applied, it’s had a lot of things thrown at it. The spec is a very thin layer on top of XML, and because of this can support a very wide range of applications. That doesn’t mean it’s necessary, or even a good idea for these purposes. But I’m afraid XOXO may be suffering from OPML envy.

It wouldn’t really be fair to criticise Attention.xml without offering a suggestion of how it might be better.

Ok, from the Attention.xml spec, here’s an example (tweaked a little to make the URIs clearer) of the per-feed part, first in OPML:

<outline text="atitle"
         description="descstr"
         url="http://aurl"
         atomurl="http://anatomurl"
         xmlurl="http://anrssurl"
         type="typestr" ...>

<li><a href="http://url" type="typestr">atitle</a>
    <dl><dt>description</dt><dd>descstr</dd>
        <dt>alturls</dt>
        <dd><ul><li><a href="http://anatomurl">atomurl</a></li>
                <li><a href="http://anrssurl">xmlurl</a></li></ul></dd>
    </dl>
</li>

So first pass, keeping the OPML structure but renaming the element to what it represents:

<feed text="atitle"
         description="descstr"
         url="http://aurl"
         atomurl="http://anatomurl"
         xmlurl="http://anrssurl" ...>

<feed url="http://aurl">
     <title>atitle</title>
     <alturls>
         <atomurl>http://anatomurl</atomurl>
          <xmlurl>http://anrssurl</xmlurl>
     </alturls>
</feed>

<feed url="http://aurl">
     <title>atitle</title>
     <alturls>
         <url href="http://anatomurl">atomurl</url>
         <url href="http://anrssurl">xmlurl</url>
     </alturls>
</feed>

<channel rdf:about="http://aurl"

      xmlns="http://purl.org/rss/1.0/"
      xmlns:atn="http://purl.org/stuff/attentionRDF"
      xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#">

     <title>atitle</title>
     <link rdf:resource="http://anatomurl" atn:format="atomurl" />
     <link rdf:resource="http://anrssurl" atn:format="xmlurl" />
</channel>

In conclusion then I’d say that the Technoratists are on to something with Attention.xml (and it’s really nice to see the bits they’ve implemented/made public so far). I think they’ve probably identified most if not all of the pieces of interest. I believe they’ve been wise avoiding the anti-interop that is OPML, and their use of XHTML outlines in XOXO is ingenious. But in the process of putting the data in XOXO they’ve lost the benefits of a custom XML language without gaining many of the advantages of a general-purpose data language (this deficit can be overcome, either at producer or consumer simply by using a mapping as in GRDDL). I doubt whether this will be an issue in practice, as the format is capable of expressing what is required. The adoption of other formats in the syndication space has had a lot more to do with marketing, blog-hype and perceived lack of alternatives than any technical merits.

Having said all that, it would be nice to have a general mapping from XOXO to RDF. So if anyone now feels that they no longer need Steve Gillmor’s $80 report, sling the dollars my way and I’ll sort out some XSLT ;-)

PS. Dare Obasanjo’s report highlights the question of the problem statement, and how much Attention.xml can help (along with some other good points).