Add normalized DataCite affiliation split

README.md

@@ -1,19 +1,23 @@
 ---
 license: cc0-1.0
 language:
-  - multilingual
+- multilingual
 pretty_name: DataCite 2026-07 Unique Affiliation Strings
 size_categories:
-  - 1M<n<10M
+- 1M<n<10M
 tags:
-  - datacite
-  - affiliations
-  - scholarly-metadata
+- datacite
+- affiliations
+- scholarly-metadata
 configs:
-  - config_name: default
-    data_files:
-      - split: train
-        path: data/*.parquet
+- config_name: default
+  data_files:
+  - split: train
+    path: data/train-*.parquet
+  - split: normalized
+    path:
+    - data/normalized-00000-of-00002.parquet
+    - data/normalized-00001-of-00002.parquet
 ---
 
 # DataCite 2026-07 Unique Affiliation Strings
@@ -53,3 +57,14 @@ print(dataset["train"][0]["affiliation"])
 ## License
 
 Released under CC0 1.0, consistent with the DataCite public data source.
+
+<!-- normalized-affiliations:start -->
+
+## Normalized affiliations
+
+The `normalized` split uses `comet-datacite-ascii-v1`: Unicode text is
+transliterated to ASCII, lowercased, stripped of punctuation while preserving
+internal whitespace, and trimmed. Empty results are excluded. Its columns are
+`normalized_affiliation` (string) and `raw_affiliation_count` (uint64).
+
+<!-- normalized-affiliations:end -->

data/normalized-00000-of-00002.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:90315de1dcd3675858e15fc016c83aec614813b9d1ced45afb065bfbecdac8c1
+size 18215146

data/normalized-00001-of-00002.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53101a2ab9f2feb5e859936a0415bca83449e0ead9dc0ada77b18b478764f6d6
+size 8743367

normalization_metadata.json

@@ -0,0 +1,62 @@
+{
+  "generated_at": "2026-08-28T07:58:32.353806+00:00",
+  "normalization": {
+    "empty_result_policy": "exclude",
+    "name": "comet-datacite-ascii-v1",
+    "steps": [
+      "transliterate Unicode text to ASCII with Unidecode",
+      "convert letters to lowercase",
+      "remove punctuation while preserving internal whitespace",
+      "trim surrounding whitespace"
+    ],
+    "unidecode_version": "1.4.0"
+  },
+  "normalizer": {
+    "script": "/data/normalize_datacite.py",
+    "sha256": "358b51fb26741c92e801264bf3ce250ecbfd278673624e62b01a10c84f9db86d"
+  },
+  "runtime": {
+    "dependencies": {
+      "PyYAML": "6.0.3",
+      "Unidecode": "1.4.0",
+      "duckdb": "1.5.5",
+      "huggingface-hub": "1.28.0",
+      "numpy": "2.3.5",
+      "pyarrow": "21.0.0"
+    },
+    "python_version": "3.12.12"
+  },
+  "schema_version": 1,
+  "shards": [
+    {
+      "bytes": 18215146,
+      "path": "data/normalized-00000-of-00002.parquet",
+      "rows": 1000000,
+      "sha256": "90315de1dcd3675858e15fc016c83aec614813b9d1ced45afb065bfbecdac8c1"
+    },
+    {
+      "bytes": 8743367,
+      "path": "data/normalized-00001-of-00002.parquet",
+      "rows": 521265,
+      "sha256": "53101a2ab9f2feb5e859936a0415bca83449e0ead9dc0ada77b18b478764f6d6"
+    }
+  ],
+  "source": {
+    "files": [
+      "data/train-00000-of-00001.parquet"
+    ],
+    "repo_id": "cometadata/datacite-unique-affiliations",
+    "revision": "613a0122b134c3c4a6dbf3615411808d0b768f82",
+    "rows": 1680322
+  },
+  "split": {
+    "excluded_raw_affiliation_count": 136,
+    "name": "normalized",
+    "represented_raw_rows": 1680186,
+    "rows": 1521265,
+    "schema": {
+      "normalized_affiliation": "string",
+      "raw_affiliation_count": "uint64"
+    }
+  }
+}