1
0
Fork 0

Added indi_funder_country_collab

Kept only indi_pub_has_cc_licence
This commit is contained in:
dimitrispie 2021-11-26 16:13:10 +02:00
parent 8750a71502
commit 09fc2afdca
1 changed files with 18 additions and 7 deletions

View File

@ -236,13 +236,13 @@ on p.id= tmp.id;
-- de facto standard in scholarly communication and is promoted by many initiatives
-- like Plan S. This indicator might be only useful when applied
-- to openly available publications.
create table indi_pub_has_cc_licence_tr stored as parquet as
select distinct p.id, case when lic='' or lic is null then 0 else 1 end as has_cc_license_tr
from publication p
left outer join (select p.id, license.type as lic from publication p
join publication_licenses as license on license.id = p.id
where lower(license.type) LIKE '%creativecommons.org%' OR lower(license.type) LIKE '%cc-%') tmp
on p.id= tmp.id
--create table indi_pub_has_cc_licence_tr stored as parquet as
--select distinct p.id, case when lic='' or lic is null then 0 else 1 end as has_cc_license_tr
--from publication p
--left outer join (select p.id, license.type as lic from publication p
--join publication_licenses as license on license.id = p.id
--where lower(license.type) LIKE '%creativecommons.org%' OR lower(license.type) LIKE '%cc-%') tmp
--on p.id= tmp.id
-- #EOSC-F2-01M_cc Rich metadata for scholarly publications
-- ## Indicator: has_cc_license. Creative Commons licensing has become a
@ -299,6 +299,17 @@ join tmp as o2 on o1.result=o2.result
where o1.id<>o2.id
group by o1.id, o2.id, o1.type
create table indi_funder_country_collab stored as parquet as
with tmp as (select funder, project, country from organization_projects op
join organization o on o.id=op.id
join project p on p.id=op.project
where country <> 'UNKNOWN')
select f1.funder, f1.country, f2.country, count(distinct f1.project) as collaborations
from tmp as f1
join tmp as f2 on f1.project=f2.project
where f1.country<>f2.country
group by f1.funder, f2.country, f1.country
create table indi_pub_diamond stored as parquet as
select distinct pd.id, coalesce(in_diamond_journal, 0) as in_diamond_journal
from publication_datasources pd