Infoscience imports documentation

General workflow : Modular Python Scripts based on tailored Python clients

The data pipeline is breaked down into separate Python scripts, each responsible for a specific task

1. Harvesting: Regularly fetch publications from Wos and Scopus based on a publication date range.
2. Deduplication: Merge the fetched data into a single dataframe with unique dedupicated metadata. Deduplicate on the existing publicatioins in Infoscience
3. Enrichment: Separate authors and affiliations into a second dataframe and enrich it with local laboratory and author informations using api.epfl.ch
4. Loading: Push the new metadata in Infoscience using the Dspace API client to create/enrich new works and persons entities.

main.py : acts as an orchestrator

Python scripts

Located in ./data_pipeline folder.

1. harvester.py: Fetch publications from different sources (Wos and Scopus for the moment). Each source is harvested with a dedicated client, one client by sources in the clients folder. They all are run in a Harvester class that can be easily extended to support multiple sources. This approach allows to separate the harvesting logic from the source-specific implementation details.

2. deduplicator.py: Merge and deduplicate the fetched data.

   The final dataframe contains following metadata :

   source : source KB (wos, scopus)

   internal_id: publication Id in the source KB (WOS:xxxx, eid)

   doi

   title

   doi

   doctype: the doctype in the source KB

   pubyear

   ifs3_doctype: the Infoscience doctype

   ifs3_collection_id: the Infoscience collection Id (depending on doctype)

   authors, and affiliations

3. enricher.py: Enrich the authors and affiliations dataframe with local laboratory information (for authors) and Unpaywall OA attributes (for publications).

4. loader.py: Push the metadata into Dspace-CRIS using the Dspace API client.

Orchestrator

main.py in ./data_pipeline folder : chains the operations.

Contains the default queries for the external sources. These default queries can be overwritten by passing new queries as parameter of the main function

For example :

df_metadata, df_authors, df_epfl_authors, df_unloaded = main(start_date="2023-01-01", end_date="2023-12-31")

Or

custom_queries = {
    "wos": "OG=(Your Custom Query for WOS)",
    "scopus": "AF-ID(Your Custom Scopus ID)",
    "openalex": "YOUR_CUSTOM_OPENALEX_QUERY",
    "zenodo": "YOUR_CUSTOM_ZENODO_QUERY"
}
df_metadata, df_authors, df_epfl_authors, df_unloaded = main(start_date="2023-01-01", end_date="2023-12-31", queries=custom_queries)

Clients

Located in ./clients folder.

Each source of metadata is harvested and parsed by a specific client, before the data being processed in the python scripts.

1. wos_client_v2.py: contains the WosClient class with all methods to parse the results of the WoS search API
2. scopus_client_v2.py: contains the ScopusClient class with all methods to parse the results of the Scopus search API
3. api_epfl_client.py : contains the ApiEpflClient class for local EPFL informations retrieval (author sciper Id, accreds and units)
4. unpaywall_client.py : contains the UnpaywallClient class with methods to request the Unpaywall API
5. dspace_client_wrapper.py: contains the DSpaceClientWrapper with methods to search and update objects in Dspace using the Dspace Rest Client
6. zenodo_client.py : contains the ZenodoClient class with all methods to parse the results of the Zenodo search API

Others : some tests with Orcid API for managing authors names.

Mappings

All mappings are in mappings.py

Internal script used (one shot) to create the mapping dictionary between Infoscience collection labels and Infoscience collection id

url = "https://infoscience.epfl.ch/server/api/core/collections"
params = {"page":0, "size": 25}
response = requests.get(url, params=params).json()
#[{"collection_uuid":x["uuid"],"entity_type":x["metadata"]["dc.title"][0]["value"]} for x in response["_embedded"]["collections"]]
collections_mapping = {}
for x in response["_embedded"]["collections"]:
    collections_mapping[x["metadata"]["dc.title"][0]["value"]] = x["uuid"]
collections_mapping

Returns

{'Patents': 'ce5a1b89-cfb3-40eb-bdd2-dcb021e755b7',
 'Projects': '49ec7e96-4645-4bc0-a015-ba4b81669bbc',
 'Teaching Materials': 'c7e018d4-2349-46dd-a8a4-c32cf5f5f9a1',
 'Images, Videos, Interactive resources, and Design': '329f8cd3-dc1a-4228-9557-b27366d71d41',
 'Newspaper, Magazine, or Blog post': '971cc7fa-b177-46e3-86a9-cfac93042e9d',
 'Funding': '8b185e36-0f99-4669-9a46-26a19d4f3eab',
 'Other': '0066acb2-d5c0-49a0-b273-581df34961cc',
 'Datasets and Code': '33a1cd32-7980-495b-a2bb-f34c478869d8',
 'Student works': '305e3dad-f918-48f6-9309-edbeb7cced14',
 'Units': 'bc85ee71-84b0-4f78-96a1-bab2c50b7ac9',
 'Contents': 'e8dea11e-a080-461b-82ee-6d9ab48404f3',
 'Virtual collections': '78f331d1-ee55-48ef-bddf-508488493c90',
 'EPFL thesis': '4af344ef-0fb2-4593-a234-78d57f3df621',
 'Reports, Documentation, and Standards': 'd5ec2987-2ee5-4754-971b-aca7ab4f9ab7',
 'Preprints and Working Papers': 'd8dada3a-c4bd-4c6f-a6d7-13f1b4564fa4',
 'Books and Book parts': '1a71fba2-2fc5-4c02-9447-f292e25ce6c1',
 'Persons': '6acf237a-90d7-43e2-82cf-c3591e50c719',
 'Events': '6e2af01f-8b92-461e-9d08-5e1961b9a97b',
 'Conferences, Workshops, Symposiums, and Seminars': 'e91ecd9f-56a2-4b2f-b7cc-f03e03d2643d',
 'Journals': '9ada82da-bb91-4414-a480-fae1a5c02d1c',
 'Journal articles': '8a8d3310-6535-4d3a-90b6-2a4428097b5b'}

Configs

Some pipeline's configurations are in config.py

• default_queries (on datasources) : queries used by the harvesters
• source_order : order of preferred item (according datasource) for deduplicate operation
• scopus_epfl_afids: list of structures Scopus Ids in Scopus referential (used for discriminate EPFL authors for Scopus datasource)
• LICENSE_CONDITIONS : dict of conditions for parsing the best_oa_location from Unpaywall

Tests and examples

Documentation on using clients and scripts : documentation_and_examples.ipynb

To test the Python scripts : demo_pipeline.ipynb

Airflow

Docker installation

docker build . -f Dockerfile-airflow --pull --tag airflow:custom
docker run -it --name=airflow -p 8081:8080 -v $PWD/clients:/opt/airflow/dags/clients  -v $PWD/data_pipeline:/opt/airflow/dags/data_pipeline  -v $PWD/imports_dag.py:/opt/airflow/dags/imports_dag.py -it airflow:custom airflow webserver
docker exec -it airflow airflow scheduler

Important : la création du user admin ne semble pas fonctionner depuis le Dockerfile, il faut exécuter en depuis le container

airflow users create --username admin --password admin --firstname Admin --lastname Admin --role Admin --email admin@example.org