OCR quality estimate for Compact Memory pages
An estimate of the OCR quality of every page of Compact Memory, and the mean per journal.
Per page (jl:ocrKnownWordShare): the share of its words (letters only, at least two characters, lowercased) that
occur at least 10 times in every 20th page of the corpus. Per journal (jl:meanOcrKnownWordShare): the mean over its
pages. An estimate, not a measurement: Hebrew, Yiddish and Russian pages score lower partly because the word list is
dominated by German. The statements are about the pages and journals of compact-memory; the texts come from the
full text of the FID portal’s Compact Memory search.
| Kind: | Complement |
|---|---|
| Method: |
Statistical
Automated: statements made by a method that can be wrong individually and were not checked one by one. |
| Maintained by: | Kai Eckert |
| License: |
CC0
|
| Generator: | generators/cm-ocr-quality (how every version was made) |
| All locations: | Where the files of this dataset are (data finder) |
Versions
2026-10-12 stable · Kai Eckert
| File | Size | Triples | SHA-256 |
|---|---|---|---|
| cm-ocr-quality.ttl.gz | 7.4 MB | 846,808 | 50de7954aa34… |
| cm-ocr-quality-meta.ttl.gz | 2.2 kB | 58 | e617aa1832a4… |
Changelog
- 2026-10-12 (version 2026-10-12), Kai Eckert: First version: the OCR quality estimate of the dashboard as a dataset (decision 0020).
