---------- Messaggio inoltrato ---------- Da: "Marco Fossati" fossati@fbk.eu Data: 11 nov 2016 1:23 PM Oggetto: Fwd: Re: [wikicite-discuss] Entity tagging and fact extraction (from a scholarly publisher perspective) A: "Marco Fossati" fossati@spaziodati.eu Cc:
---------- Messaggio inoltrato ---------- Da: "Marco Fossati" fossati@fbk.eu Data: 11 nov 2016 1:18 PM Oggetto: Re: [wikicite-discuss] Entity tagging and fact extraction (from a scholarly publisher perspective) A: "Andrew Smeall" andrew.smeall@hindawi.com Cc: "Dario Taraborelli" dtaraborelli@wikimedia.org, "Benjamin Good" < ben.mcgee.good@gmail.com>, "Discussion list for the Wikidata project." < wikidata@lists.wikimedia.org>, "wikicite-discuss" < wikicite-discuss@wikimedia.org>, "Daniel Mietchen" < daniel.mietchen@googlemail.com>
Hi everyone,
Just a couple of thoughts, which are in line with Dario's first message: 1. the primary sources tool lets third party providers release *full datasets* in a rather quick way. It is conceived to (a) ease the ingestion of *non-curated* data and to (b) make the community directly decide which statements should be included, instead of eventually complex a priori discussions. Important: the datasets should comply with the Wikidata vocabulary/ontology.
2. I see the mix'n'match tool as a way to *link* datasets with Wikidata via ID mappings, thus only requiring statements that say "Wikidata entity X links to the third party dataset entity Y". This is pretty much what the linked data community has been doing so far. No need to comply with the Wikidata vocabulary/ontology.
Best,
Marco
Il 11 nov 2016 10:27 AM, "Andrew Smeall" andrew.smeall@hindawi.com ha scritto:
Regarding the topics/vocabularies issue:
A challenge we're working on is finding a set of controlled vocabularies for all the subject areas we cover.
We do use MeSH for those subjects, but this only applies to about 40% of our papers. In Engineering, for example, we've had more trouble finding an open taxonomy with the same level of depth as MeSH. For most internal applications, we need 100% coverage of all subjects.
Machine learning for concept tagging is trendy now, partly because it doesn't require a preset vocabulary, but we are somewhat against this approach because we want to control the mapping of terms and a taxonomic hierarchy can be useful. The current ML tools I've seen can match to a controlled vocabulary, but then they need the publisher to supply the terms.
The temptation to build a new vocabulary is strong, because it's the fastest way to get to something that is non-proprietary and universal. We can merge existing open vocabularies like MeSH and PLOS to get most of the way there, but we then need to extend that with concepts from our corpus.
Thanks Daniel and Benjamin for your responses. Any other feedback would be great, and I'm always happy to delve into issues from the publisher perspective if that can be helpful.
On Fri, Nov 11, 2016 at 4:54 PM, Dario Taraborelli < dtaraborelli@wikimedia.org> wrote:
Benjamin – agreed, I too see Wikidata as mainly a place to hold all the mappings. Once we support federated queries in WDQS, the benefit of ID mapping (over extensive data ingestion) will become even more apparent.
Hope Andrew and other interested parties can pick up this thread.
On Wed, Nov 2, 2016 at 12:11 PM, Benjamin Good ben.mcgee.good@gmail.com wrote:
Dario,
One message you can send is that they can and should use existing controlled vocabularies and ontologies to construct the metadata they want to share. For example, MeSH descriptors would be a good way for them to organize the 'primary topic' assertions for their articles and would make it easy to find the corresponding items in Wikidata when uploading. Our group will be continuing to expand coverage of identifiers and concepts from vocabularies like that in Wikidata - and any help there from publishers would be appreciated!
My view here is that Wikidata can be a bridge to the terminologies and datasets that live outside it - not really a replacement for them. So, if they have good practices about using shared vocabularies already, it should (eventually) be relatively easy to move relevant assertions into the WIkidata graph while maintaining interoperability and integration with external software systems.
-Ben
On Wed, Nov 2, 2016 at 8:31 AM, 'Daniel Mietchen' via wikicite-discuss < wikicite-discuss@wikimedia.org> wrote:
I'm traveling ( https://twitter.com/EvoMRI/status/793736211009536000 ), so just in brief: In terms of markup, some general comments are in https://www.ncbi.nlm.nih.gov/books/NBK159964/ , which is not specific to Hindawi but partly applies to them too.
A problem specific to Hindawi (cf. https://commons.wikimedia.org/wiki/Category:Media_from_Hindawi) is the bundling of the descriptions of all supplementary files, which translates into uploads like https://commons.wikimedia.org/wiki/File:Evolution-of-Coronar y-Flow-in-an-Experimental-Slow-Flow-Model-in-Swines-Angiogra phic-and-623986.f1.ogv (with descriptions for nine files) and eight files with no description, e.g. https://commons.wikimedia.org/wiki/File:Evolution-of-Coronar y-Flow-in-an-Experimental-Slow-Flow-Model-in-Swines-Angiogra phic-and-623986.f2.ogv .
There are other problems in their JATS, and it would be good if they would participate in http://jats4r.org/ . Happy to dig deeper with Andrew or whoever is interested.
Where they are ahead of the curve is licensing information, so they could help us set up workflows to get that info into Wikidata.
In terms of triple suggestions to Wikidata:
- as long as article metadata is concerned, I would prefer to
concentrate on integrating our workflows with the major repositories of metadata, to which publishers are already posting. They could help us by using more identifiers (e.g. for authors, affiliations, funders etc.), potentially even from Wikidata (e.g. for keywords/ P921, for both journals and articles) and by contributing to the development of tools (e.g. a bot that goes through the CrossRef database every day and creates Wikidata items for newly published papers).
- if they have ways to extract statements from their publication
corpus, it would be good if they would let us/ ContentMine/ StrepHit etc. know, so we could discuss how to move this forward. d.
On Wed, Nov 2, 2016 at 1:42 PM, Dario Taraborelli dtaraborelli@wikimedia.org wrote:
I'm at the Crossref LIVE 16 event in London where I just gave a
presentation
on WikiCite and Wikidata targeted at scholarly publishers.
Beside Crossref and Datacite people, I talked to a bunch of folks
interested
in collaborating on Wikidata integration, particularly from PLOS,
Hindawi
and Springer Nature. I started an interesting discussion with Andrew
Smeall,
who runs strategic projects at Hindawi, and I wanted to open it up to everyone on the lists.
Andrew asked me if – aside from efforts like ContentMine and StrepHit
–
there are any recommendations for publishers (especially OA
publishers) to
mark up their contents and facilitate information extraction and
entity
matching or even push triples to Wikidata to be considered for
ingestion.
I don't think we have a recommended workflow for data providers for facilitating triple suggestions to Wikidata, other than leveraging the Primary Sources Tool. However, aligning keywords and terms with the corresponding Wikidata items via ID mapping sounds like a good first
step. I
pointed Andrew to Mix'n'Match as a handy way of mapping identifiers,
but if
you have other ideas on how to best support 2-way integration of
Wikidata
with scholarly contents, please chime in.
Dario
--
Dario Taraborelli Head of Research, Wikimedia Foundation wikimediafoundation.org • nitens.org • @readermeter
-- WikiCite 2016 – May 26-26, 2016, Berlin Meta: https://meta.wikimedia.org/wiki/WikiCite_2016 Twitter: https://twitter.com/wikicite16
You received this message because you are subscribed to the Google
Groups
"wikicite-discuss" group. To unsubscribe from this group and stop receiving emails from it,
send an
email to wikicite-discuss+unsubscribe@wikimedia.org.
-- WikiCite 2016 – May 26-26, 2016, Berlin Meta: https://meta.wikimedia.org/wiki/WikiCite_2016 Twitter: https://twitter.com/wikicite16
You received this message because you are subscribed to the Google Groups "wikicite-discuss" group. To unsubscribe from this group and stop receiving emails from it, send an email to wikicite-discuss+unsubscribe@wikimedia.org.
--
*Dario Taraborelli *Head of Research, Wikimedia Foundation wikimediafoundation.org • nitens.org • @readermeter http://twitter.com/readermeter
-- WikiCite 2016 – May 26-26, 2016, Berlin Meta: https://meta.wikimedia.org/wiki/WikiCite_2016 Twitter: https://twitter.com/wikicite16
You received this message because you are subscribed to the Google Groups "wikicite-discuss" group. To unsubscribe from this group and stop receiving emails from it, send an email to wikicite-discuss+unsubscribe@wikimedia.org.
--
Andrew Smeall Head of Strategic Projects
Hindawi Publishing Corporation Kirkman House 12-14 Whitfield Street, 3rd Floor London, W1T 2RF United Kingdom
-- WikiCite 2016 – May 26-26, 2016, Berlin Meta: https://meta.wikimedia.org/wiki/WikiCite_2016 Twitter: https://twitter.com/wikicite16
You received this message because you are subscribed to the Google Groups "wikicite-discuss" group. To unsubscribe from this group and stop receiving emails from it, send an email to wikicite-discuss+unsubscribe@wikimedia.org.