Upload 1M deduped paper text dataset
Browse files- README.md +72 -0
- stats.json +45 -0
- train_00000.parquet +3 -0
- train_00001.parquet +3 -0
- train_00002.parquet +3 -0
- train_00003.parquet +3 -0
- train_00004.parquet +3 -0
- train_00005.parquet +3 -0
- train_00006.parquet +3 -0
- train_00007.parquet +3 -0
- train_00008.parquet +3 -0
- train_00009.parquet +3 -0
README.md
ADDED
|
@@ -0,0 +1,72 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
pretty_name: Deduped Paper Text Dataset
|
| 3 |
+
viewer: true
|
| 4 |
+
tags:
|
| 5 |
+
- datasets
|
| 6 |
+
- arxiv
|
| 7 |
+
- scientific-papers
|
| 8 |
+
- text
|
| 9 |
+
size_categories:
|
| 10 |
+
- 1M<n<10M
|
| 11 |
+
---
|
| 12 |
+
|
| 13 |
+
# Deduped Paper Text Dataset
|
| 14 |
+
|
| 15 |
+
This dataset merges an existing paper-text parquet export with one or more
|
| 16 |
+
backfill parquet shards, then keeps exactly one row per `canonical_paper_id`.
|
| 17 |
+
|
| 18 |
+
One row corresponds to one canonical arXiv paper. Where multiple versions are
|
| 19 |
+
available, the merge keeps the strongest text row according to the repo's
|
| 20 |
+
selection logic: prefer non-partial text, then prefer raw-PDF-derived text,
|
| 21 |
+
then newer versions, then longer text.
|
| 22 |
+
|
| 23 |
+
## Important provenance note
|
| 24 |
+
|
| 25 |
+
This is a merged release, not a fresh one-pass export. It combines an earlier
|
| 26 |
+
deduped paper-text dataset with later backfill shards, then dedupes again on
|
| 27 |
+
`canonical_paper_id`.
|
| 28 |
+
|
| 29 |
+
Use per-paper `license` filtering before broad downstream publication. Many
|
| 30 |
+
arXiv records still carry `nonexclusive-distrib` rather than a general reuse
|
| 31 |
+
license.
|
| 32 |
+
|
| 33 |
+
This release is capped deterministically at exactly `1000000` deduped papers. The selected inputs contained `1000004` deduped papers before applying the cap.
|
| 34 |
+
|
| 35 |
+
## Rows
|
| 36 |
+
|
| 37 |
+
- `train`: `1000000` papers
|
| 38 |
+
- unique canonical papers: `1000000`
|
| 39 |
+
- base input rows considered: `124036`
|
| 40 |
+
- backfill input rows considered: `875968`
|
| 41 |
+
- deduped rows before target cap: `1000004`
|
| 42 |
+
|
| 43 |
+
|
| 44 |
+
## Main columns
|
| 45 |
+
|
| 46 |
+
- `paper_id`
|
| 47 |
+
- `canonical_paper_id`
|
| 48 |
+
- `paper_version`
|
| 49 |
+
- `pdf_path`
|
| 50 |
+
- `title`
|
| 51 |
+
- `abstract`
|
| 52 |
+
- `authors`
|
| 53 |
+
- `categories`
|
| 54 |
+
- `license`
|
| 55 |
+
- `text`
|
| 56 |
+
- `text_char_count`
|
| 57 |
+
- `token_count`
|
| 58 |
+
- `page_count`
|
| 59 |
+
- `token_types`
|
| 60 |
+
|
| 61 |
+
## Files
|
| 62 |
+
|
| 63 |
+
- `train_00000.parquet`
|
| 64 |
+
- `train_00001.parquet`
|
| 65 |
+
- `train_00002.parquet`
|
| 66 |
+
- `train_00003.parquet`
|
| 67 |
+
- `train_00004.parquet`
|
| 68 |
+
- `train_00005.parquet`
|
| 69 |
+
- `train_00006.parquet`
|
| 70 |
+
- `train_00007.parquet`
|
| 71 |
+
- `train_00008.parquet`
|
| 72 |
+
- `train_00009.parquet`
|
stats.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"base_parquets": [
|
| 3 |
+
"/arxiv/huggingface/paper_text_124k_dedup_v1/train_00000.parquet",
|
| 4 |
+
"/arxiv/huggingface/paper_text_124k_dedup_v1/train_00001.parquet"
|
| 5 |
+
],
|
| 6 |
+
"backfill_parquets": [
|
| 7 |
+
"/arxiv/pdfs_structured_gcs_growth_v1/paper_text_backfill_00000.parquet",
|
| 8 |
+
"/arxiv/pdfs_structured_gcs_growth_v1/paper_text_backfill_00001.parquet",
|
| 9 |
+
"/arxiv/pdfs_structured_gcs_growth_v1/paper_text_backfill_00002.parquet",
|
| 10 |
+
"/arxiv/pdfs_structured_gcs_growth_v1/paper_text_backfill_00003.parquet",
|
| 11 |
+
"/arxiv/pdfs_structured_gcs_growth_v1/paper_text_backfill_00004.parquet",
|
| 12 |
+
"/arxiv/pdfs_structured_gcs_growth_v1/paper_text_backfill_00005.parquet",
|
| 13 |
+
"/arxiv/pdfs_structured_gcs_growth_v1/paper_text_backfill_00006.parquet",
|
| 14 |
+
"/arxiv/pdfs_structured_gcs_growth_v1/paper_text_backfill_00007.parquet",
|
| 15 |
+
"/arxiv/pdfs_structured_gcs_growth_v1/paper_text_backfill_00008.parquet",
|
| 16 |
+
"/arxiv/pdfs_structured_gcs_growth_v1/paper_text_backfill_00009.parquet",
|
| 17 |
+
"/arxiv/pdfs_structured_gcs_growth_v1/paper_text_backfill_00010.parquet",
|
| 18 |
+
"/arxiv/pdfs_structured_gcs_growth_v1/paper_text_backfill_00011.parquet"
|
| 19 |
+
],
|
| 20 |
+
"metadata_path": "/data/arxiv/arxiv-metadata-oai-snapshot.json",
|
| 21 |
+
"output_dir": "/arxiv/huggingface/paper_text_1m_dedup_v1",
|
| 22 |
+
"base_rows": 124036,
|
| 23 |
+
"backfill_rows": 875968,
|
| 24 |
+
"merged_rows": 1000000,
|
| 25 |
+
"merged_unique_canonical_ids": 1000000,
|
| 26 |
+
"target_rows": 1000000,
|
| 27 |
+
"merged_rows_before_target_cap": 1000004,
|
| 28 |
+
"merged_unique_canonical_ids_before_target_cap": 1000004,
|
| 29 |
+
"metadata_subset_rows": 1000000,
|
| 30 |
+
"rows_per_output_file": 100000,
|
| 31 |
+
"memory_limit": "16GB",
|
| 32 |
+
"output_files": [
|
| 33 |
+
"/arxiv/huggingface/paper_text_1m_dedup_v1/train_00000.parquet",
|
| 34 |
+
"/arxiv/huggingface/paper_text_1m_dedup_v1/train_00001.parquet",
|
| 35 |
+
"/arxiv/huggingface/paper_text_1m_dedup_v1/train_00002.parquet",
|
| 36 |
+
"/arxiv/huggingface/paper_text_1m_dedup_v1/train_00003.parquet",
|
| 37 |
+
"/arxiv/huggingface/paper_text_1m_dedup_v1/train_00004.parquet",
|
| 38 |
+
"/arxiv/huggingface/paper_text_1m_dedup_v1/train_00005.parquet",
|
| 39 |
+
"/arxiv/huggingface/paper_text_1m_dedup_v1/train_00006.parquet",
|
| 40 |
+
"/arxiv/huggingface/paper_text_1m_dedup_v1/train_00007.parquet",
|
| 41 |
+
"/arxiv/huggingface/paper_text_1m_dedup_v1/train_00008.parquet",
|
| 42 |
+
"/arxiv/huggingface/paper_text_1m_dedup_v1/train_00009.parquet"
|
| 43 |
+
],
|
| 44 |
+
"readme_path": "/arxiv/huggingface/paper_text_1m_dedup_v1/README.md"
|
| 45 |
+
}
|
train_00000.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ecbb422f2e4fe3c14f3755983d7ecabfb617b967cbea586d474accddba479e80
|
| 3 |
+
size 2328555934
|
train_00001.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0684a0ca0f9e015d166d47f5423d3c0fb565016c1e86f7c694fdeb840206f7d8
|
| 3 |
+
size 2273234539
|
train_00002.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c4417b1f5f5586bfb1780912eca271bf510c8b62a163213f54cacea47191aebb
|
| 3 |
+
size 2161849469
|
train_00003.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c7cf981a44ec50ae5c1795220a2a5792441e57ce93bf5dcdc4f129605f95033e
|
| 3 |
+
size 2073528823
|
train_00004.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:fcc74d6a0ab52ee879158c2de5232c86b187ad081f0a960c34819823b88546c7
|
| 3 |
+
size 2160875026
|
train_00005.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0cdbd3349bae90fe233cf80c06cbb4c23e6706555e3fec179458e17f9ba9a4c6
|
| 3 |
+
size 2300787241
|
train_00006.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2f4826f5b342ba59fa79badac3ba1203565cf1d8b6d24707ebf89cf2fdb4b193
|
| 3 |
+
size 2196641190
|
train_00007.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3a9fb0ab77ab6a0e699c9b6ebc12205edf4d67f2df8766c7a4e7202d6991c2fb
|
| 3 |
+
size 2292537056
|
train_00008.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7bd406b77221e3dcadc189178d0ca5338b1a9e073f2aefbbace579c945b8469b
|
| 3 |
+
size 1938291752
|
train_00009.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:1d90dafa099b05bdc9b289a4c389602ff940b67bad047e1f48b2d076fa911b18
|
| 3 |
+
size 2034683832
|