Works not affiliated to university

[Ivo Bleylevens]

Dear community,

I have a list of +-23.000 items that are registered as local output of our university in our own CRIS, but looking at those items in OpenAlex, they are not registered as our output.

I would like to update this, but what is the best way to do this ? Send an email to support or through https://works-magnet.esr.gouv.fr (I dont know how to operate there), or is this only available for OpenAlex Premium subscribers ?

After correcting this I hope OpenAlex also get better in linking items to our university that are not yet stored in our local CRIS. 

Looking forward to hear your experiences !

Kind regards,
Ivo

Dr. Ivo Bleylevens 
Developer / Research Intelligence / Data science
 Maastricht University Library 
www.maastrichtuniversity.nl
Grote Looiersstraat 17, 6211 JH Maastricht
T 043-3885029 
https://orcid.org/0000-0003-3884-9280
https://www.linkedin.com/in/ivobleylevens

[Gabor Schubert]

Dear Ivo,

I made similar observations with Stockholm University (SU) local data vs OpenAlex data. I tested all the ca. 48000 publications with DOI (2012-2024) which are registered in our local repository and affiliated to our university. 8000 of these are included in OpenAlex but not affiliated with SU there, and roughly half of these 8000 have no affiliation data at all in OpenAlex. As far as I understand, the Works-magnet can be helpful to identify affiliations for publications which have raw affiliation data, but the algorithm was not able to connect these to ROR-affiliations, and I'm not sure if the manual feedback/curation requests can work with thousands of publications. As far as I see missing affiliation data is more common for "non-journal article" type publications, which usually have less structured original metadata in sources OpenAlex harvests its data.

On the other hand, I found quite a lot of publications which are affiliated with SU according to OpenAlex but are not registered in our local repository. These are mainly "non-journal article" type publications (for example: datasets, preprints), which are usually not registered locally as publications, but I also found quite a few publications which are marked as journal articles, but they were meeting abstracts or preprints in reality (which are not registered locally).

Gabor Schubert
Stockholm University

[Eric Jeangirard]

Dear Ivo,

Could you share a few examples of those works ?

Cheers

Eric

--
You received this message because you are subscribed to the Google Groups "OpenAlex Community" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openalex-commun...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/openalex-community/131da806-ca0f-4d32-80c2-802805e36ea0n%40googlegroups.com.

[Gabor Schubert]

Hi,

There is a way to estimate the number and ratio of publications in OpenAlex which have no affiliation data at all: the value "institutions_distinct_count" (docs.openalex.org/api-entities/works/work-object#institutions_distinct_count) shows the number of institutions which are matched to a certain publication. One can use a filter to show the publications (and the number of publications) which have 0 institutions listed as affiliations, for example the following command: (api.openalex.org/works?filter=type:article,institutions_distinct_count:0,publication_year:2024) will show the number of publications from 2024 which are articles and have no affiliated institution data.

There are ca 270 million publications in OpenAlex currently and ca 160 million of these have no affiliation data according to the "institutions_distinct_count" filter. Apparently, this is more common for non-article type publications where only ca. 30% of the publications have matched affiliation data. So obviously there are tens of millions of publications in OpenAlex which have no affiliation data at all, thus they won't be shown when one searches for affiliations. Most probably these are from sources which have lower quality data/metadata.

Gabor Schubert
Stockholm University

[Subbiah Arunachalam]

Friends,

It is a serious issue if a large proportion of works are not assigned affiliations. I am sure folks at OpenAlex would set things right very soon.

Best wishes.

Subbiah Arunachalam

[Gabor Schubert]

Hi,

There is a way to estimate the number and ratio of publications in OpenAlex which have no affiliation data at all: the value "institutions_distinct_count" (https://docs.openalex.org/api-entities/works/work-object#institutions_distinct_count) shows the number of institutions which are matched to a certain publication. One can use a filter to show the publications (and the number of publications) which have 0 institutions listed as affiliations, for example the following command https://api.openalex.org/works?filter=type:article,institutions_distinct_count:0,publication_year:2024 will show the number of publications from 2024 which are articles and have no affiliated institution data. I ran a yearly check for these for the last 10 years and got the following:
 

There are ca 270 million publications in OpenAlex currently and ca 160 million of these have no affiliation data according to the "institutions_distinct_count" filter. Apparently, this is more common for non-article type publications where only 30-40% of the publications have matched affiliations in the last 10 years. So obviously there are tens of millions of publications in OpenAlex which have no affiliation data at all, thus they won't be shown when one searches for affiliations. Most probably these are from sources which have lower quality data/metadata.

Gabor Schubert

Stockholm University

[Kevin McCurley]

It could simply be that publishers are not reporting the data when they register for a DOI. Here is an example from Springer Lecture Notes in Computer Science: https://api.crossref.org/works/10.1007/978-3-031-68309-1_8 Noticed that there are no affiliations but the HTML page for the work contains affiliations.

Kevin McCurley

[Gabor Schubert]

Hi Kevin,

You're right, probably most of these missing affiliation data in OpenAlex is caused by the lack of data in Crossref. Although other databases (like Scopus or Web of Science) are able to extract the affiliation data from other sources than Crossref (most likely from publisher data, or web scraping, or similar). I understand that this is a resource-intensive task, but the relatively large portion of missing affiliation data is a disadvantage if someone will use OpenAlex for affiliation based analysis.

Gabor Schubert

Stockholm University

[Koen Bokern]

Be sure to contact Open Alex support or Eric from Works Magnet (in the discussion above) to figure out the best approach.

I frequently use works magnet since all of our limited number of works required corrections for the matches (maybe due to ROR ID registration after matching model training or due to issues with generic name elements such as university of applied sciences) .

It may not be the best approach, but I queried your organization in works magnet and I saw that the accuracy of your existing matches seems to be quite high. For this number of works the results pages are not responsive with my device when querying all years or a few years.

To gain insight: these works are likely to have an affiliation string related to your university, but are not matched to it:

https://openalex.org/works?page=1&filter=raw_affiliation_strings.search:maastricht+university,authorships.institutions.lineage:!i34352273

https://openalex.org/works?page=1&filter=raw_affiliation_strings.search:universiteit+maastricht,authorships.institutions.lineage:!i34352273

The matches for your close partner Maastricht University Medical Center seem to be less accurate.

They are often matched with the ROR ID: https://ror.org/036pt7h44 'University Medical Center' in Lubbock, Texas, USA as seen in works magnet.

This organization is related to Texas Tech University and Texas Tech University Health Sciences Center.

This is their OpenAlex page: https://openalex.org/institutions/i2802849423

They seem to have a very high number of collaborations with the Netherlands: https://openalex.org/works?page=1&filter=authorships.institutions.lineage:i2802849423&group_by=publication_year,open_access.is_oa,primary_topic.id,authorships.institutions.lineage,type,authorships.author.id,authorships.countries

And about 3000 works with a raw affiliation string containing 'Maastricht': https://openalex.org/works?page=1&filter=authorships.institutions.lineage:i2802849423,raw_affiliation_strings.search:maastricht&group_by=publication_year,open_access.is_oa,primary_topic.id,authorships.institutions.lineage,type,authorships.author.id,authorships.countries

Not matched for Maastricht University Medical Center:

Using 'center': https://openalex.org/works?page=1&filter=raw_affiliation_strings.search:maastricht+university+medical+center,authorships.institutions.lineage:!i2800191616

Using 'centre': https://openalex.org/works?page=1&filter=raw_affiliation_strings.search:maastricht+university+medical+centre,authorships.institutions.lineage:!i2800191616

Possible alternative strings: Maastricht Universitair Medisch Centrum, Maastricht UMC, MUMC.

The Maastricht UMC ROR ID shows 'centre': https://ror.org/02d9ce178, but since both variants do have correct matches as well this may not be an issue.

In your own interest it could be wise to request a ROR update for this US organization yourselves to prevent future matching issues, but if that organization's generic name is accurate they may not be able to make their name more specific.

Instead OpenAlex would need to adapt their matching algorithm to account for generic name elements like university medical center, or in our case university of applied sciences or they could use country info from affiliation strings to enhance matching.

In our case we were often matched with Avans University of Applied Sciences

In this case the Maastricht University Medical Center is matched with University Medical Center (the Texas Tech University affiliated one).

Those involved (like the ROR team) have stated that affiliation matching is not perfect. Therefore, it is nice to have tools like works magnet, but you could be right that it may not facilitate a very large number of corrections due to manual selection and resolving and page responsiveness.

Even if Works Magnet is not the right approach, try a query to see what it can do and what filters it provides.

If the scale of the problem would not be an issue (due to manual curation, resolving, and page responsiveness) you can query strings for the MUMC and filter on the erroneously matched US organization, then select all works after a brief check and change the ROR match in bulk.

Kind regards,

Koen Bokern, data steward at Aeres University of Applied Sciences.