Dataset Viewer
Auto-converted to Parquet Duplicate
id
string
publication_identifier
string
url
string
handle
string
doi
string
title
string
year
int64
language
string
license
string
pdf
pdf
text
string
698566db-b621-4af2-b33e-819d589eae33
019caf99b4ad-9e673450-75d2-4725-97d1-e2fe336f9bc5
https://nva.sikt.no/registration/019caf99b4ad-9e673450-75d2-4725-97d1-e2fe336f9bc5
https://hdl.handle.net/11250/5484519
null
Hav og samfunn i Norge: Nasjonale behov for havforståelse for en styrket og motstandsdyktig fremtid / Ocean & Society in Norway: National Ocean Literacy Needs for an Empowered and Resilient Future
2,025
nob
CC0
# **Hav og samfunn i Norge | Policy-notat** **Havforståelse (ocean literacy) for en styrket og motstandsdyktig fremtid** **92% opplever at de har noe kunnskap** som kan bidra til å løse utfordringer deres lokalsamfunn står overfor. **90% er villig til å engasjere seg.** Yngre folk er mindre villig. **80% opplever at ...
af3ef2f6-39af-4c93-b1f8-58f6a9783fa9
019caf99b4ad-9e673450-75d2-4725-97d1-e2fe336f9bc5
https://nva.sikt.no/registration/019caf99b4ad-9e673450-75d2-4725-97d1-e2fe336f9bc5
https://hdl.handle.net/11250/5484519
null
Hav og samfunn i Norge: Nasjonale behov for havforståelse for en styrket og motstandsdyktig fremtid / Ocean & Society in Norway: National Ocean Literacy Needs for an Empowered and Resilient Future
2,025
nob
CC0
# **Ocean & Society in Norway | Policy Brief** **National Ocean Literacy Needs for an Empowered and Resilient Future** **92% feel they have at least some knowledge** they need to overcome challenges their community face. **90% are willing to take action.** Younger people are less willing. **80% feel they have access ...
38b5f2d0-8b30-4a0c-aadb-7bfa6582ea79
019c627080f2-13b06ee7-d9b8-4f7a-8b12-5b94405a9a33
https://nva.sikt.no/registration/019c627080f2-13b06ee7-d9b8-4f7a-8b12-5b94405a9a33
https://hdl.handle.net/11250/5366949
null
Effektiv ressursbruk, arbeidsglede og bærekraft: funn fra 12 intensivenheter i Norge
2,026
nob
CC0
**09:00-10:00 Registrering, kaffe og mingling 10:00-10:15 Velkommen 11:30-12:00 Pause med besøk hos utstillere 14:00-14:20 Pause og besøke utstillere 14:20-15:15 Workshop på RETHOS, NAS, og Intensivstudenters kompetanse innen mekanisk ventilering 11:00-1130 Intensivsykepleieres senkarriere: Resultater fra en nasjonal k...
2e628ee1-d60a-4e2b-8349-69cb0b30e185
019c1dd8980e-401dc454-797e-4615-9071-32ebb0e6f41e
https://nva.sikt.no/registration/019c1dd8980e-401dc454-797e-4615-9071-32ebb0e6f41e
https://hdl.handle.net/11250/5360383
null
Ramnes kirke Arkeologisk overvåking av graving for utskifting av kum
2,021
nob
CC0
# **RAMNES KIRKE** Arkeologisk overvåking av graving for utskifting av kum Halvorsen, Sunniva Wilberg #### Sammendrag NIKU overvåket tømming av vannkum 25.5.2021. Kummen ble tømt ved håndgraving, og det viste seg at det ikke ble nødvendig å fjerne kummen. Den arkeologiske overvåkingen ble deretter avsluttet. Etters...
08c30d7d-5e27-4d6b-86f4-e91f638dc541
019bc204e509-ed820e11-cd56-4fe3-9229-5156ae9afba2
https://nva.sikt.no/registration/019bc204e509-ed820e11-cd56-4fe3-9229-5156ae9afba2
https://hdl.handle.net/11250/5344855
null
Effekter av Nudging på Avfallssortering
2,025
nob
CC0
Avfalls- og gjenvinningsbransjen skal gå fra å være en del av den lineære økonomien, der ressurser utvinnes til produksjon av varer og avfall plasseres i deponi, til å bli en sirkulær økonomi som kjennetegnes ved at færrest mulig ressurser går tapt, og innebærer gjenbruk, forbedring og reparasjon (Avfall\_ Norge, 2016)...
996afa1c-9d32-4caa-a712-6b048a2f6081
019a921b5f57-97e3bb62-93b4-4e93-a493-01597d3f1776
https://nva.sikt.no/registration/019a921b5f57-97e3bb62-93b4-4e93-a493-01597d3f1776
https://hdl.handle.net/11250/5320264
null
"Klassisk musikk i grunnskolen? En diskursanalytisk studie av læreres forståelser av klassisk musi(...TRUNCATED)
2,025
nob
CC0
"# Klassisk musikk i grunnskolen?\n\nEn diskursanalytisk studie av læreres forståelser av klassisk(...TRUNCATED)
f47e4453-0ddd-4da6-a38e-5b410cd711ce
01994d06e93e-609e530d-44d9-42a6-9b87-b9d6da150898
https://nva.sikt.no/registration/01994d06e93e-609e530d-44d9-42a6-9b87-b9d6da150898
https://hdl.handle.net/11250/2374774
null
Virksomme klimabudskap
2,016
nob
CC0
"## REPORT 2016:01\n\nVirksomme klimabudskap Rapport for Naturvernforbundet\n\n### CICERO Report 201(...TRUNCATED)
622b8bbb-0354-4959-9298-8ef2cc632ae5
01990a0ecaed-14f96a7c-ebe9-4617-8972-f25ff9044120
https://nva.sikt.no/registration/01990a0ecaed-14f96a7c-ebe9-4617-8972-f25ff9044120
https://hdl.handle.net/11250/294093
null
En studie av fuktinnhold i massivtre : oppfuktings- og uttørkingsprosessen
2,015
nob
CC0
"# <span id=\"page-2-0\"></span>**Forord**\n\nDenne masteroppgaven er skrevet våren 2015 og er et s(...TRUNCATED)
44b8569c-126a-4f62-911c-2034230880fb
01990a072d68-17538a3c-8b31-4c24-9ef8-eed781b3343b
https://nva.sikt.no/registration/01990a072d68-17538a3c-8b31-4c24-9ef8-eed781b3343b
https://hdl.handle.net/11250/219260
null
Elevens opplevelse av mening og relevans i skogbruksundervisningen
2,014
nob
CC0
"## <span id=\"page-1-0\"></span>**Forord**\n\nDenne masteroppgaven har hatt som hovedformål å få(...TRUNCATED)
434cb840-9cac-4694-b54a-fcd4e80e81f2
019909f7b26d-385f1477-8c9a-428b-b0a9-d8b18e2c3cf1
https://nva.sikt.no/registration/019909f7b26d-385f1477-8c9a-428b-b0a9-d8b18e2c3cf1
https://hdl.handle.net/11250/295041
null
Demonterbar løfteramme for offshoreoperasjoner
2,015
nob
CC0
"# Demonterbar løfteramme for offshoreoperasjoner\n\nAv\n\nHenrik Lundeby Grimstad\n\nMastergradsop(...TRUNCATED)
End of preview. Expand in Data Studio

Nasjonalt vitenarkiv

Open-access documents from NVA (Nasjonalt vitenarkiv), the joint national repository where Norwegian research institutions publish their output: master's and PhD theses, journal articles, and technical and research reports. Subjects span the disciplines - marine science, forestry, archaeology, education, public health, engineering - and most documents are recent.

Each row is one PDF: the original file exactly as published, the text extracted from it, and the metadata NVA records for the publication it belongs to. Keeping the PDFs alongside the text is the point of this dataset - text extraction from PDFs is lossy and improves as tooling improves, so anyone can redo it here without re-fetching several hundred gigabytes from NVA.

Documents on NVA carry a license per file. This dataset does not filter or split on license; it keeps every openly-licensed document together and records the license per row.

Structure

One row per PDF:

column description
id NVA's identifier for the file. Unique; the key for this dataset.
publication_identifier NVA's identifier for the publication the file belongs to. Not unique - a publication can carry several PDFs.
url Landing page for the publication on nva.sikt.no.
handle Persistent handle identifier. Present for effectively every publication.
doi DOI, where the publication has one. Often null.
title Title of the publication. Not unique, for the same reason publication_identifier isn't.
year Publication year as recorded by NVA.
language nob (Bokmål), nno (Nynorsk) or nor (unspecified Norwegian).
license The file's own license: CC-BY, CC-SA or CC0.
pdf The original PDF. Renders in the dataset viewer.
text Text extracted from the PDF, as markdown.

Scope and coverage

Currently limited to Norwegian-language publications, as the dataset exists to support Norwegian language modelling. NVA also holds a large body of English-language research; including it would be a matter of widening one constant in src/nasjonalt_vitenarkiv/config.py.

Two properties of the NVA search API limit coverage, and neither is worked around fully:

  • It only pages through the first 10,000 hits of a single query, so the larger license buckets are not exhaustively covered.
  • Repeated identical queries return different subsets of the same result set. Windows of results intermittently fail with a server error in a way that does not clear on retry, and are eventually dropped. In one measurement, two consecutive queries reporting the same 404 total hits returned 97 documents each but overlapped on only 78.

Because of the second point the candidate list is treated as cumulative: each query is merged into the cached list rather than replacing it, so documents seen once are not lost when a later query misses them. A practical consequence is that coverage grows as the ingest is re-run, and that the dataset may retain a document NVA's search no longer returns.

How the text was extracted

Most of these are born-digital PDFs with a good embedded text layer, for which running a vision model over every page is wasted effort - on sampled documents it was roughly an order of magnitude slower and no more accurate. Extraction therefore runs in two passes, both using marker and both emitting markdown:

  1. Text layer, every document. Reads the PDF's embedded text, with lightweight CPU models for reading order and table structure. No vision model sees the page images.
  2. OCR, only where pass 1 found nothing usable. Documents whose extracted text is implausibly sparse for their page count (under 200 characters per page) are treated as scans and re-run through marker's full OCR path, a vision-language model over rendered pages.

So the large majority of this text is extracted rather than recognised, and is faithful to the source. The scanned minority carries the usual OCR error profile.

Limitations

Reading order can be imperfect in documents with complex layouts, and tables and equations are reconstructed on a best-effort basis. In sampling, documents laid out as forms or questionnaires were the weakest case, sometimes yielding sections in an order that does not match the page.

Some documents carry very little running text regardless of extraction quality - map sheets and figure-only appendices, for instance, reduce to little more than coordinate labels. Nothing is filtered out on those grounds here, since the aim is to mirror what NVA holds; consumers who want prose should filter on text length.

Research publications routinely contain English abstracts, references and quotations even when the body text is Norwegian.

Reproducing and updating

create.py performs the whole ingest - querying NVA, downloading PDFs, extracting text and publishing. Re-running it carries over what is already published and only processes what is new:

uv run create.py

The OCR fallback needs a GPU, Docker, and membership of the docker group, since marker runs its OCR model in a containerised vLLM server. The text-layer pass needs neither.

License

There is no single license for this dataset. See the license column for each document's license as recorded by NVA, and url or handle for the source record and its attribution requirements.

Related

The Norwegian Dynaword project publishes single-license subsets of this data, derived from it by filtering on the license column, as nva-cc0 and its siblings in Norwegian Dynaword.

Notice and takedown policy

We redistribute files shared with us under a license permitting such redistribution. If you have concerns about the licensing of these files, please contact us. If you consider that the data contains material that infringes your copyright, please:

  • Clearly identify yourself with detailed contact information such as an address, a telephone number, or an email address at which you can be contacted.
  • Clearly reference the original work claimed to be infringed.
  • Clearly identify the material claimed to be infringing and information reasonably sufficient to allow us to locate the material.

You can contact us through this channel. We will comply with legitimate requests by removing the affected documents from the next release of the dataset.


Danish Foundation Models dataset

Downloads last month
103