ACTION-282 = Draft a finding on metadata architecture

Possible areas of focus

"Metadata" is a huge, amorphous topic and it is important that we attempt to focus on a single problem or problem area. I have identified the following possibilities:

1. Null hypothesis: Metadata is just like any other kind of data, such as melting points of chemical substances. Data is of great architectural interest to W3C. However, the correct treatment of metadata is just a special case of the correct treatment of data, so metadata per se is of no special interest to W3C.
2. Data about documents: We define metadata in the traditional sense used in information science: it is descriptive information about document-like things. We abuse the word "document" to construe it very broadly - so not just books and articles but also static web pages, audio recordings, images, video, and other similar resources. Metadata architecture then applies to the creation, maintenance, transmission, and application of this particular kind of information. This is of interest to W3C because so much of the value of the Web is tied up in this sort of object. [Note: JAR doesn't like stretching words like "document" in this way, but this term was adopted as a compromise in a discussion on this subject.]
3. Data about Web plumbing: Many discussions about metadata drift away from data about documents and into adjacent ontological territory involved with making the web work. For example, data about people, organizations, hosts, network services, and other dynamic resources is interesting because all of these entities participate in creating the value of the web - the interest goes beyond their role in creating and consuming documents. (Examples: WSDL, XRD.) While it is correct to say that such data is not metadata, doing so does not satisfy the thirst for a uniform treatment of such plumbing-related information.

For the next draft (in progress), I propose looking at the both of the latter two, eventually either dropping one or, if necessary, splitting into two documents.

An architectural document is most successful when it starts from problems and applications, not technologies. If we look at the protocols and formats that are often discussed in the context of metadata (XML, OWL, POWDER, Link: header, and so on), we find that many of them are applicable to any of the above concentration areas (data-generally, data-about-documents, data-about-web-plumbing), and this fact helps explain why discussions can get so confusing. I propose that for the next draft discussion of any technologies or other considerations that seem to be independent of application area be relegated to a late section or appendix to help make its independence clear.

TBD: (1) Organize this document into three parts (a) data-about-documents (b) data-about-web-plumbing (c) common technological base. (2) For each format or technology mentioned in "What deployments / use cases are inspirational?" below, and in the earlier survey, provide a link or snippet that concretely illustrates the format, and locate one or more user-facing applications that make use of that format. E.g. for Dublin Core, find some actual "wild type" metadata that uses the DC vocabulary, and find at least one application that puts DC metadata to some interesting use (indexing, browsing, etc.).

Why should the TAG get into this?

Why should the TAG care about metadata?

• it's the TAG's job to encourage a connected, open, inclusive Web
• there's a vague feeling that there might be something to be learned by looking at metadata globally...
• in particular, something that might help us understand and evaluate some of the many puzzles that come our way...
  • RDFa vs. microdata
  • LRDD / Link: / .well-known / <link>
  • XRD
  • HTTP semantics (including redirections)
  • using URIs to "refer" vs. to "locate"
  • <link rel="canonical">
  • multimedia bookmarking
  • what is "authoritative"
• and also might help us identify opportunities for advocacy and/or standardization

What is the division of responsibility between TAG and others (e.g. W3C working groups)?

• WGs have looked at various pieces (RDF, SKOS) but not the big picture. That's as it should be
• interests of WGs may not be aligned with those of the TAG (for which see above)
• heavy existing interest in many communities (e.g. digital repositories)
• IETF mostly takes care of relevant protocols

What could we possibly do? We're not a WG.

• Descriptive text: Here is the current state of the art, presented in an organized, thoughtful way; with gap analysis.
• Prescriptive text: Here are some ways in which, hypothetically, things might be made better through the actions of WGs, Web publishers, and others.
• Facilitation (?): ask so-and-so to talk to so-and-so.

Questions

For further research:

Does metadata have any special role on the web (as compared to other kinds of content)?

Yes, metadata makes the data (and therefore the Web) more valuable in particular ways that both promote interoperability and benefit from it.

• makes things more discoverable both locally and globally
• enables management of collections of things
• opens up communication channels (e.g. provenance, licensing)

Is there a problem? In what ways is metadata on the web less webby (connected, open, inclusive) than it ought to be?

Not enough of it (example?) - poor incentives for creating it

• most web pages don't have it
• <author> was a failure, value not realized
• why bother with <link> or RDFa ?
• level of reuse of metadata is inadequate to justify investment
• authors not usually competent or motivated to curate their own stuff
• creator / curator / consumer interests not aligned

Difficult to deploy

• well, maybe no easier or harder than web publishing generally
• structured formats are always more painful than free text
• LRDD will only be used when someone really needs to use it
• RDFa is early days, relatively untested, poor tool support

Hard to validate

A lot of what's there is closed, e.g.

• openurl DOI metadata sits behind login
• Pubmed can be accessed piecemeal, RESTfully, but bulk load requires license
• iTunes ? (research this)

Difficult to use at scale

• many many gardens: indexes, repositories, databases, catalogs (Pubmed, Citeulike, IMDB, Flickr, Amazon, ...)
• nothing web-like (other than big search engines)

Being a master consumer of metadata is complicated (XMP, GIF, <link>, LRDD, 303, RDFa, ...)

• not everyone can do this
• too many "standard" reference formats
• too many protocols (examples?)

Doubt and uncertainty regarding data identity

• what exactly is it about - what's the subject?
• is this about that, or not?
• when are two metadatums about the same data?

Unclear lines of authority, thus difficult to evaluate for trustworthiness (but how is this different from any other content on the web?)

How to consistently identify people, organizations, places (organize bookmarks by author, photos by place)

What metadata do we care about?

Is there such a thing as data that is not metadata? Metadata that is not data? Can non-data have metadata? If X is about Y, does that mean X is in scope of this project, or is X only in scope when Y is data?

Are OpenID and XRD a metadata use case, or just application-related data?

Is RDF nose-following (linked data) a metadata use case, a world unto itself, or an intersecting world?

Should we focus our attention on particular metadata profiles (e.g. Dublin Core), or on the meta-metadata problem (bibliographic would be a special case, offers to sell might another, audio another, etc.)?

• standardizing particular things is always warm and fuzzy, and there's a huge need for agreed-upon formats...
• probably that should be done in a WG, not by the TAG

Is "metadata" even the right word to cover this project?

What deployments / use cases are inspirational?

Both private and public, that is. Figure out business models, especially for the more public sources.

• CiteSeer
• Delicious
• iTunes
• Papers
• Flickr ?
• Creative Commons licensing
• how is XMP metadata actually used?
• Google Scholar ??
• Tabulator ???
• any cool SKOS clients? (check recent thread)
• Gene Ontology annotations

What potential technical opportunities are there?

Things someone might do in RDF-land to advance web metadata:

• applicability statements - what to use, when, how
• common representations - particular vocabularies, ontologies, schemas
• style recommendations
• metadata deployment guide(s)
• richer logical models, with relations between entities (e.g. part/whole, versions, quality)
• ontologies of metadata subjects (class hierarchy)
• subject headings (SKOS)
• how to serialize using an RDF serialization
• how to import RDF into non-RDF settings (example please?)
• why is RDF, which was designed for this purpose, not getting uptake in this application area? Is there something someone can do to make it a better fit to community needs?

Elsewhere:

• what would we have to say to someone who is not touching RDF?

Does "metadata architecture" make sense? Is it something to be discovered (empirical), designed (invented), or some of each?

Is anything different now compared to 10 years ago when RDF and Dublin Core were published?

The Metadata Activity Statement (1998) is worth a look.

• more experience, more frustration
• W3C recommendations haven't met with universal acclaim
• How do they fall short? are they fixable?
• Does any "market" (community) care enough to do anything? Is there will?