Datasets:
Vasily Alexeev commited on
Commit ·
71433f1
1
Parent(s): 6927795
Selected -> Thematized
Browse files
README.md
CHANGED
|
@@ -15,7 +15,7 @@ configs:
|
|
| 15 |
path: data/topics.jsonl.gz
|
| 16 |
---
|
| 17 |
|
| 18 |
-
# Russian Wikipedia quality articles with subjects
|
| 19 |
|
| 20 |
This Russian-language collection contains **17,251 Wikipedia articles**
|
| 21 |
with at least one subject label from a fixed taxonomy of **28 subjects**.
|
|
@@ -83,7 +83,7 @@ Of the **17,413 source articles**, **14,945** required no additional review:
|
|
| 83 |
**2,468** articles had evidence from **44 ambiguous editorial rubrics** and
|
| 84 |
received additional AI-assisted review. Of these, **2,306** were retained and
|
| 85 |
**162** were excluded because their final subject lists were empty. In total,
|
| 86 |
-
**2,054** articles had their subject lists changed. The final
|
| 87 |
contains **17,251 articles: 15,599 with one subject and 1,652 with several**.
|
| 88 |
Additional review records a step in annotation: it does not imply an automatic
|
| 89 |
label change and is not a confidence score. Articles without additional review
|
|
@@ -181,7 +181,7 @@ The repository does not store a second permanent copy as VW text files.
|
|
| 181 |
The NLP preparation section describes the normalized namespaces and text fields.
|
| 182 |
|
| 183 |
The sidecar [data/labels.json](data/labels.json) is a JSON object mapping every
|
| 184 |
-
|
| 185 |
and the CSV `@topic` namespace. It is not a separate dataset configuration.
|
| 186 |
|
| 187 |
```python
|
|
@@ -195,7 +195,7 @@ labels = json.loads(Path("data/labels.json").read_text(encoding="utf-8"))
|
|
| 195 |
|
| 196 |
[data/topic-provenance.jsonl.gz](data/topic-provenance.jsonl.gz) contains one
|
| 197 |
record for each of the **17,413 Full article IDs**, including the 162 articles
|
| 198 |
-
excluded from
|
| 199 |
annotation metadata, separate from corpus reconstruction and NLP build outputs;
|
| 200 |
preprocessing does not re-annotate it or create new subject evidence. It records
|
| 201 |
the initial specific-rubric assignments and the final accepted assignments,
|
|
@@ -205,11 +205,11 @@ without publishing internal annotation decision logs.
|
|
| 205 |
| --- | --- | --- |
|
| 206 |
| `id` | string | Full article ID, `ruwiki/<page_id>`; join to Full to inspect excluded articles. |
|
| 207 |
| `initial_topics` | array of strings | Subjects supported by specific direct editorial rubrics before additional review. |
|
| 208 |
-
| `topics` | array of strings | Final subjects; matches `data/labels.json` for every
|
| 209 |
| `additional_review` | boolean | Whether baseline ambiguous-rubric evidence led to additional AI-assisted review; not whether labels changed or a confidence score. |
|
| 210 |
| `unresolved_rubrics` | array of strings | Ambiguous direct editorial rubrics at the initial stage; retained as historical evidence even after review, not a claim that the final subjects remain unresolved. |
|
| 211 |
|
| 212 |
-
To stream only reviewed articles retained in
|
| 213 |
the downloaded repository. Use `not row["topics"]` instead of `row["topics"]`
|
| 214 |
to inspect the excluded articles, joining their IDs to Full:
|
| 215 |
|
|
@@ -358,7 +358,7 @@ can inspect these losses or apply a different extraction procedure. Neither
|
|
| 358 |
The [standalone preparation code](https://huggingface.co/datasets/TopicNet/RuWiki-Good-2026/tree/main/preprocessing)
|
| 359 |
and [full instructions](https://huggingface.co/datasets/TopicNet/RuWiki-Good-2026#nlp-preparation)
|
| 360 |
are maintained in Full. Run the preparation commands from Full's
|
| 361 |
-
`preprocessing/` directory, including when preparing only
|
| 362 |
repository directories may have any names. Replace the paths below with your own
|
| 363 |
absolute paths: `work_dir` is a separate working directory, and `selected_data`
|
| 364 |
points to this collection's `data/` directory. Choose fresh output and scratch
|
|
@@ -400,7 +400,7 @@ for a new build. Full's instructions explain reuse of local JSONL files and
|
|
| 400 |
access to restricted source repositories.
|
| 401 |
|
| 402 |
After the structural NLP checks, the shared entry point independently compares
|
| 403 |
-
CSV and
|
| 404 |
in the corpus contract, reporting `PUBLISHED_NLP_OK`. It does not rely on the
|
| 405 |
new build manifest as its authority and does not replay NLP normalization.
|
| 406 |
|
|
@@ -429,7 +429,7 @@ progress; the manifest reports empty `@word` articles by count and ordered IDs
|
|
| 429 |
without dropping them.
|
| 430 |
|
| 431 |
Bigrams and trigrams use independent NLTK PMI selection after frequency ≥5,
|
| 432 |
-
with `nbest(32100)[100:]` for each order. This selection uses **all
|
| 433 |
article bodies**, independently of Full; it is corpus-global rather than fitted
|
| 434 |
only on a future training partition. For text-only experiments, use only `@word`
|
| 435 |
to avoid exposing category, title or subject metadata to the model. The workflow
|
|
@@ -451,12 +451,12 @@ bow = load_dataset(
|
|
| 451 |
)
|
| 452 |
```
|
| 453 |
|
| 454 |
-
A measured NLP build of **both Full and
|
| 455 |
42 min 5.83 s, with 10.58 GiB peak resident memory, approximately 3.74 GiB of final
|
| 456 |
output and 0.53 GiB of scratch. These are measurements from one environment,
|
| 457 |
-
not
|
| 458 |
|
| 459 |
-
## Rebuild
|
| 460 |
|
| 461 |
Full's [corpus replay instructions](https://huggingface.co/datasets/TopicNet/RuWiki-Good-2026#rebuild-the-corpus-from-wikimedia-dumps)
|
| 462 |
cover the six pinned 20261001 dumps, explicit download resumption with `--resume`,
|
|
@@ -489,7 +489,7 @@ annotation directory must already contain this collection's `labels.json` and
|
|
| 489 |
`topics.jsonl.gz`. Reconstruction retains the published mapping and does not
|
| 490 |
rerun AI annotation or establish new evidence for label correctness.
|
| 491 |
|
| 492 |
-
If a previously accepted Full corpus is already available, build only
|
| 493 |
without rereading the dumps. From the same Full `preprocessing/` directory,
|
| 494 |
resolve Full's data path:
|
| 495 |
|
|
@@ -506,7 +506,7 @@ python prepare.py --stage corpus --action verify --dataset selected \
|
|
| 506 |
--output-dir "$work_dir/selected-output"
|
| 507 |
```
|
| 508 |
|
| 509 |
-
This
|
| 510 |
archive of all 17,251 articles, the labels file and the topic definitions.
|
| 511 |
|
| 512 |
Use fresh output and scratch directories. Corpus verification checks exact
|
|
|
|
| 15 |
path: data/topics.jsonl.gz
|
| 16 |
---
|
| 17 |
|
| 18 |
+
# RuWiki-Good-Thematized-2026 — Russian Wikipedia quality articles with subjects
|
| 19 |
|
| 20 |
This Russian-language collection contains **17,251 Wikipedia articles**
|
| 21 |
with at least one subject label from a fixed taxonomy of **28 subjects**.
|
|
|
|
| 83 |
**2,468** articles had evidence from **44 ambiguous editorial rubrics** and
|
| 84 |
received additional AI-assisted review. Of these, **2,306** were retained and
|
| 85 |
**162** were excluded because their final subject lists were empty. In total,
|
| 86 |
+
**2,054** articles had their subject lists changed. The final Thematized collection
|
| 87 |
contains **17,251 articles: 15,599 with one subject and 1,652 with several**.
|
| 88 |
Additional review records a step in annotation: it does not imply an automatic
|
| 89 |
label change and is not a confidence score. Articles without additional review
|
|
|
|
| 181 |
The NLP preparation section describes the normalized namespaces and text fields.
|
| 182 |
|
| 183 |
The sidecar [data/labels.json](data/labels.json) is a JSON object mapping every
|
| 184 |
+
Thematized article ID to its list of subject slugs, matching both article `topics`
|
| 185 |
and the CSV `@topic` namespace. It is not a separate dataset configuration.
|
| 186 |
|
| 187 |
```python
|
|
|
|
| 195 |
|
| 196 |
[data/topic-provenance.jsonl.gz](data/topic-provenance.jsonl.gz) contains one
|
| 197 |
record for each of the **17,413 Full article IDs**, including the 162 articles
|
| 198 |
+
excluded from Thematized. Records are ordered by numeric page ID. This is frozen
|
| 199 |
annotation metadata, separate from corpus reconstruction and NLP build outputs;
|
| 200 |
preprocessing does not re-annotate it or create new subject evidence. It records
|
| 201 |
the initial specific-rubric assignments and the final accepted assignments,
|
|
|
|
| 205 |
| --- | --- | --- |
|
| 206 |
| `id` | string | Full article ID, `ruwiki/<page_id>`; join to Full to inspect excluded articles. |
|
| 207 |
| `initial_topics` | array of strings | Subjects supported by specific direct editorial rubrics before additional review. |
|
| 208 |
+
| `topics` | array of strings | Final subjects; matches `data/labels.json` for every Thematized ID. An empty list identifies an excluded Full article. |
|
| 209 |
| `additional_review` | boolean | Whether baseline ambiguous-rubric evidence led to additional AI-assisted review; not whether labels changed or a confidence score. |
|
| 210 |
| `unresolved_rubrics` | array of strings | Ambiguous direct editorial rubrics at the initial stage; retained as historical evidence even after review, not a claim that the final subjects remain unresolved. |
|
| 211 |
|
| 212 |
+
To stream only reviewed articles retained in Thematized, run this from the root of
|
| 213 |
the downloaded repository. Use `not row["topics"]` instead of `row["topics"]`
|
| 214 |
to inspect the excluded articles, joining their IDs to Full:
|
| 215 |
|
|
|
|
| 358 |
The [standalone preparation code](https://huggingface.co/datasets/TopicNet/RuWiki-Good-2026/tree/main/preprocessing)
|
| 359 |
and [full instructions](https://huggingface.co/datasets/TopicNet/RuWiki-Good-2026#nlp-preparation)
|
| 360 |
are maintained in Full. Run the preparation commands from Full's
|
| 361 |
+
`preprocessing/` directory, including when preparing only Thematized. Local
|
| 362 |
repository directories may have any names. Replace the paths below with your own
|
| 363 |
absolute paths: `work_dir` is a separate working directory, and `selected_data`
|
| 364 |
points to this collection's `data/` directory. Choose fresh output and scratch
|
|
|
|
| 400 |
access to restricted source repositories.
|
| 401 |
|
| 402 |
After the structural NLP checks, the shared entry point independently compares
|
| 403 |
+
CSV and Thematized label byte sizes and SHA-256 hashes with published expectations
|
| 404 |
in the corpus contract, reporting `PUBLISHED_NLP_OK`. It does not rely on the
|
| 405 |
new build manifest as its authority and does not replay NLP normalization.
|
| 406 |
|
|
|
|
| 429 |
without dropping them.
|
| 430 |
|
| 431 |
Bigrams and trigrams use independent NLTK PMI selection after frequency ≥5,
|
| 432 |
+
with `nbest(32100)[100:]` for each order. This selection uses **all Thematized
|
| 433 |
article bodies**, independently of Full; it is corpus-global rather than fitted
|
| 434 |
only on a future training partition. For text-only experiments, use only `@word`
|
| 435 |
to avoid exposing category, title or subject metadata to the model. The workflow
|
|
|
|
| 451 |
)
|
| 452 |
```
|
| 453 |
|
| 454 |
+
A measured NLP build of **both Full and Thematized** from ready JSONL snapshots took
|
| 455 |
42 min 5.83 s, with 10.58 GiB peak resident memory, approximately 3.74 GiB of final
|
| 456 |
output and 0.53 GiB of scratch. These are measurements from one environment,
|
| 457 |
+
not Thematized-only estimates or measurements of replay from raw Wikimedia dumps.
|
| 458 |
|
| 459 |
+
## Rebuild Thematized from source data
|
| 460 |
|
| 461 |
Full's [corpus replay instructions](https://huggingface.co/datasets/TopicNet/RuWiki-Good-2026#rebuild-the-corpus-from-wikimedia-dumps)
|
| 462 |
cover the six pinned 20261001 dumps, explicit download resumption with `--resume`,
|
|
|
|
| 489 |
`topics.jsonl.gz`. Reconstruction retains the published mapping and does not
|
| 490 |
rerun AI annotation or establish new evidence for label correctness.
|
| 491 |
|
| 492 |
+
If a previously accepted Full corpus is already available, build only Thematized
|
| 493 |
without rereading the dumps. From the same Full `preprocessing/` directory,
|
| 494 |
resolve Full's data path:
|
| 495 |
|
|
|
|
| 506 |
--output-dir "$work_dir/selected-output"
|
| 507 |
```
|
| 508 |
|
| 509 |
+
This Thematized-only workflow has been verified byte for byte against the published
|
| 510 |
archive of all 17,251 articles, the labels file and the topic definitions.
|
| 511 |
|
| 512 |
Use fresh output and scratch directories. Corpus verification checks exact
|