Compact Memory in figures
The Jewish periodicals digitised by the UB Frankfurt, page by page: what the full text contains, how readable the OCR is, where the source data has gaps, and how the pages are linked into the JudaicaLink knowledge graph.
Pages over time
Pages with full text per year of issue. Hover a year to see which journals make up the volume and how readable the OCR is.
Pages per year
Average OCR quality per year
share of known words (cm-ocr-quality); years with at least 50 pagesShow the numbers as a table
Compact Memory in the graph
The pages, issues and journals are records of the JudaicaLink graph (dataset compact-memory). In 2021, TagMe linked mentions on the pages to entities (cm-tagme); each entity is represented by an entity page. Every number here is a SPARQL query on the platform's endpoint: follow “query” to run it.
Share of pages with a linked entity, per year
Entities on most pages
Journals by linked share
Pages with at least one mention / pages of the journal; journals with 1,000 pages or more.
The journals with the lowest share
Journals
The 40 largest journals by decade, then every journal of the catalogue. Select a journal for its profile and a sample page.
The 40 largest journals, pages per decade
| Notes |
|---|
What the pages are
Language comes from the journal's catalogue record, the unit from the node a page hangs under in the Visual Library (issue, article, volume).
Language of the journal
Unit the page belongs to
OCR quality
Pages by share of known words
Journals with the weakest OCR
At least 500 pages. Hebrew, Yiddish and Russian score lower partly because the word list is mostly German.
Data issues
What was left out of the full text and what is wrong in the source metadata. Pages were dropped, not repaired; a fresh export from the UB Frankfurt should fix most of this at the source.
From catalogue to full text
Metadata coverage of the pages
Queries
The SPARQL queries behind the figures, with the time they took on the platform's Fuseki when this page was made.