RE: The Agent proposal in bib.schema.org is controversial

MARC 720 is the main culprit. Here’s the analysis Richard referred to is below:

Jeff

---

Many of the Agents coming out of the transform process are produced from the MARC 720 field where the 1st indicator is blank (“Not specified”) or 2 (“Other”):

http://www.loc.gov/marc/bibliographic/bd720.html


Note that the MARC 720 field is used when crosswalking Dublin Core to MARC:

http://www.loc.gov/marc/dccross.html


Here’s a quote:

“Note: there is no way to specify whether the Contributor is a person or organization because it is not in the Dublin Core data. If it can reasonably be determined that the contributor is a person or organization, fields 700 1#$a (Added Entry--Personal Name) or 710 2#$a (Added Entry--Corporate Name) may be used.”
In other words, anyone who wants to upgrade their Dublin Core data to Schema.org will have this same problem when trying to map dc:creator, dc:contributor, dct:mediator, and dct:rightsHolder, not to mention dc:subject.

In some cases, our transformer is able to sniff the name’s structure to choose Person or Organization. (BTW, there is still room for some improvement there.) Without such clues, though, the process assigns bgn:Agent as a default, based on the assumption that reconciliation with typed entities will be done downstream.

I can give plenty of examples, but here’s one to illustrate:
(The highlighted rdf:value statements are used for debugging and contain the source data using in the mapping.)

<http://www.worldcat.org/oclc/881301071> # "我的動物小百科"
    schema:contributor<http://schema.org/contributor> <http://experiment.worldcat.org/entity/work/data/1913968092#Agent/ke_xue_guan> ; # "科學館"
    .
<http://experiment.worldcat.org/entity/work/data/1913968092#Agent/ke_xue_guan> # "科學館"
    a bgn:Agent<http://bibliograph.net/Agent> ;
    rdfs:label<http://www.w3.org/2000/01/rdf-schema#label> "Ke xue guan" ;
    schema:name<http://schema.org/name> "科學館"@zh ;
    rdf:value<http://www.w3.org/1999/02/22-rdf-syntax-ns#value> "<v720 altscript=\".//v880[16]\" i1=\" \" i2=\" \"><s6><d>880-16</d></s6><sa><d>Ke xue guan.</d></sa></v720>"^^rdf:XMLLiteral ; # idiomatic diagnostic
    rdf:value<http://www.w3.org/1999/02/22-rdf-syntax-ns#value> "<v880 i1=\" \" i2=\" \" lnkfrom=\"720\" script=\"$1\" xlink=\".//v720[1]\"><s6><d>720-16/$1</d></s6><sa><d>科學館</d></sa></v880>"^^rdf:XMLLiteral ; # idiomatic diagnostic
    .
Jeff

From: Corey A Harper [mailto:corey.harper@nyu.edu]
Sent: Monday, August 10, 2015 3:11 PM
To: Dan Scott
Cc: Young,Jeff (OR); LeVan,Ralph; Richard Wallis; public-schemabibex@w3.org
Subject: Re: The Agent proposal in bib.schema.org is controversial

Dear all,

My $0.02: I also think that schama:Thing is the best option at this time, and don't think we should push too much on Agent given what I consider relatively limited usefulness. I understand Jeff's point about the dangers of "not sorting these out", but I also think that we can store and manage data with whatever specifity we want, and I'm not sure those dangers apply to data as published downstream to consumers on the Web.

I'm also _very_ interested in knowing more about the 70 Million + "mystery agents" Richard and Jeff have been referencing. Are these just 1xx and 7xx data points that are type unknown because they haven't matched a known entity with a known type? Can't we at least infer more about their type by their Marc field? Can we see some example instance (bib) data where these show up?

Best,
-Corey

On Mon, Aug 10, 2015 at 1:55 PM, Dan Scott <denials@gmail.com<mailto:denials@gmail.com>> wrote:
FWIW, the Bibliographic Ontology (bibo) also uses foaf:Agent.

But I concur with the developing dissenting opinion on the github issue that, if we have nothing specific to say about the nature of the entity because we lack the information, it's better to simply avoid the compromise of Agent. We might make ourselves feel a bit better about the dismal state of our bibliographic data through an abstract class like Agent, but in the end it doesn't really add any data to the data we're trying to express.

Using schema:Thing seems like an acceptable fallback in the mean time, and allows the data expressed by the target links to be refined to either Person or Organization at some point in the future when the effort occurs.


On Mon, 10 Aug 2015 at 11:29 Young,Jeff (OR) <jyoung@oclc.org<mailto:jyoung@oclc.org>> wrote:
I made an argument that the problem is broader than bib records:

https://github.com/schemaorg/schemaorg/issues/700#issuecomment-129078302


Limiting to our situation, though, Richard cites the count from WorldCat at 72 million “agents” (people and organizations excluded):

https://github.com/schemaorg/schemaorg/issues/700#issuecomment-129227478


These all have Linked Data identifiers, but they are only mechanized placeholders in need of exposure, reconciliation, and enrichment.

The danger of not sorting these out is that naïve automated “entity matching” processes resort to string matching on name as an “else condition” and the resulting mix up manifests itself in the Linked Data.

I suggested Google Custom Search as a possible tool to help with discovery and possibly lead to an interface where they could be reconciled:

https://github.com/schemaorg/schemaorg/issues/700#issuecomment-129239474


Jeff

From: LeVan,Ralph
Sent: Monday, August 10, 2015 10:33 AM
To: Young,Jeff (OR); Richard Wallis; public-schemabibex@w3.org<mailto:public-schemabibex@w3.org>

Subject: RE: The Agent proposal in bib.schema.org<http://bib.schema.org> is controversial

One of the arguments against Agent was that if you didn’t know what kind of object a thing was, then you just shouldn’t say.   All the properties of Agent seem to come from Thing.  I’d propose that we just use Thing.

My guess is that the need for Agent comes mostly from our need to convert existing bib records into RDF and some of our crappy old bib records don’t reliably distinguish the type of agent involved.  Rather than be caught out in a lie about whether the agent is a Person or Organization, we’d rather say less.  This is a problem peculiar to our situation and not a broad problem of the internet community.  It’s also a short-term problem.  Selling ‘Agent’ to a community that doesn’t need it is going to be an uphill battle.

What’s wrong with dropping all the way back to Thing when we don’t know the type of the agent?

Ralph

From: Young,Jeff (OR) [mailto:jyoung@oclc.org]
Sent: Monday, August 10, 2015 10:04 AM
To: Richard Wallis; public-schemabibex@w3.org<mailto:public-schemabibex@w3.org>
Subject: RE: The Agent proposal in bib.schema.org<http://bib.schema.org> is controversial

One option would be for us to use foaf:Agent. Presumably search engines would ignore it, but that’s their prerogative.

Another option would be to preserve http://bibliograph.net/Agent, with a comment that it wasn’t accepted by the broader community, but remains useful in our limited domain. (Terms that have been adopted should be deprecated.)

Jeff


From: Richard Wallis [mailto:richard.wallis@dataliberate.com]
Sent: Monday, August 10, 2015 8:18 AM
To: public-schemabibex@w3.org<mailto:public-schemabibex@w3.org>
Subject: The Agent proposal in bib.schema.org<http://bib.schema.org> is controversial

You may have noticed if you followed the recent announcement of Schema.or v2.1<https://lists.w3.org/Archives/Public/public-schemabibex/2015Aug/0000.html>, which includes bib.schema.org<http://bib.schema.org>, that one of our proposals did not make it in.  That proposal being the Agent type that we proposed as a super-type for Person and Organization.

Agent has been a theme of discussion in the community well before we approached the issue.  You can follow the recent debate in the related schemaorg git issue comment trail: https://github.com/schemaorg/schemaorg/issues/700


In the bibliographic world Agent is a well understood, some would say obvious, approach.  When applied to the wider domains that Schema.org embraces however, it raises many concerns and issues. Especially because, as proposed, it would introduce a new direct sub-type of Thing with ramifications that could cascade across many areas of the  vocabulary.

In my personal opinion the gap between the two apposing views on this is significant and the best way forward would be to consider possible pragmatic approaches to how we represent our data in Schema.org without loosing the ability to describe our resources effectively to the wider world.

In simple terms, if we identify an author, creator, publisher, or even copyright holder as a Person or an Organization there is not a problem.  The difficulty occurs when we know from the relationships in the data that they are either a Person or an Organization but cannot identify which.

One suggested way forward for such a circumstance would be to define them as a schema:Thing.  To me this feels a little too vague.  A follow-on option was to suggest a 'personOrOrganization' boolean property to indicate this circumstance.  This is a little more appealing, but I think it still needs some work.

What are others thoughts on this?

Do we believe that the proposed Agent type is the only way forward?  Are there potential pragmatic options like the one I describe above that we could shape, that would be acceptable? Is this requirement to specifically describe agents as too detailed and something we can pass over, and move on to other things?

~Richard.



Richard Wallis
Founder, Data Liberate
http://dataliberate.com

Linkedin: http://www.linkedin.com/in/richardwallis

Twitter: @rjw



--
Corey A Harper
Metadata Services Librarian
New York University Libraries
20 Cooper Square, 3rd Floor
New York, NY 10003-7112
212.998.2479
corey.harper@nyu.edu<mailto:corey.harper@nyu.edu>