Upload folder using huggingface_hub

README.md

@@ -46,14 +46,21 @@ Each record contains:
 |-------|------|-------------|
 | `arxiv_doi` | string | The DOI for the arXiv work (format: `10.48550/arXiv.{id}`) |
 | `arxiv_id` | string | The arXiv identifier (e.g., `1412.6980`) |
-| `citation_count` | integer | Number of citing works |
-| `cited_by` | array | List of citation objects |
+| `reference_count` | integer | Total number of reference instances |
+| `citation_count` | integer | Number of unique citing works |
+| `cited_by` | array | List of citing work objects |
 
-Each citation object in `cited_by` contains:
+Each citing work object in `cited_by` contains:
 
 | Field | Type | Description |
 |-------|------|-------------|
 | `doi` | string | DOI of the citing work |
+| `matches` | array | List of reference matches from this citing work |
+
+Each match object in `matches` contains:
+
+| Field | Type | Description |
+|-------|------|-------------|
 | `raw_match` | string | The original text that matched the arXiv reference |
 | `reference` | string | The full reference object from the citing work's metadata (JSON-encoded) |
 
@@ -63,12 +70,17 @@ Each citation object in `cited_by` contains:
 {
   "arxiv_doi": "10.48550/arXiv.1412.6980",
   "arxiv_id": "1412.6980",
+  "reference_count": 40040,
   "citation_count": 40040,
   "cited_by": [
     {
       "doi": "10.1016/j.jhydrol.2025.134709",
-      "raw_match": "arXiv.1412.6980",
-      "reference": "{\"key\": \"10.1016/j.jhydrol.2025.134709_b0175\", \"unstructured\": \"Kingma, D.P., & Ba, J. (2014), Adam: A Method for Stochastic Optimization...\"}"
+      "matches": [
+        {
+          "raw_match": "arXiv.1412.6980",
+          "reference": "{\"key\": \"10.1016/j.jhydrol.2025.134709_b0175\", \"unstructured\": \"Kingma, D.P., & Ba, J. (2014), Adam: A Method for Stochastic Optimization...\"}"
+        }
+      ]
     }
   ]
 }
@@ -77,7 +89,7 @@ Each citation object in `cited_by` contains:
 To parse the reference field in Python:
 ```python
 import json
-reference = json.loads(record['cited_by'][0]['reference'])
+reference = json.loads(record['cited_by'][0]['matches'][0]['reference'])
 ```
 
 ## Extraction Process

data/arxiv_citations_valid.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9a585b3818eef324918e238004fbd4b9efc7d5ff3f4e7e9d8d4809e89fd4a2fb
-size 464962508
+oid sha256:e8b0fb3b96b64ed854d860ccfda676426c8a52c32c361c2f938b810e362479d8
+size 458932570

transform_schema.py

@@ -0,0 +1,133 @@
+#!/usr/bin/env python3
+"""
+Transform existing arxiv_citations parquet to new grouped schema.
+
+Old schema:
+    cited_by: [{doi, raw_match, reference}, ...]
+
+New schema:
+    reference_count: total reference instances
+    citation_count: unique citing works
+    cited_by: [{doi, matches: [{raw_match, reference}, ...]}, ...]
+"""
+
+import json
+from collections import defaultdict
+from pathlib import Path
+
+import pyarrow as pa
+import pyarrow.parquet as pq
+
+
+def transform_record(record: dict) -> dict:
+    """Transform a single record to the new grouped schema."""
+    # Group citations by citing DOI
+    citing_works = defaultdict(list)
+
+    for citation in record.get('cited_by', []):
+        doi = citation.get('doi', '')
+        match_entry = {
+            'raw_match': citation.get('raw_match', ''),
+            'reference': citation.get('reference', ''),  # Already JSON string
+        }
+        citing_works[doi].append(match_entry)
+
+    # Convert to new format
+    cited_by = [
+        {'doi': doi, 'matches': matches}
+        for doi, matches in citing_works.items()
+    ]
+
+    return {
+        'arxiv_doi': record.get('arxiv_doi', ''),
+        'arxiv_id': record.get('arxiv_id', ''),
+        'reference_count': sum(len(cw['matches']) for cw in cited_by),
+        'citation_count': len(cited_by),
+        'cited_by': cited_by,
+    }
+
+
+def main():
+    input_path = Path('data/arxiv_citations_valid.parquet')
+    output_path = Path('data/arxiv_citations_valid_v2.parquet')
+
+    print(f"Reading {input_path}...")
+    table = pq.read_table(input_path)
+    records = table.to_pylist()
+    print(f"Loaded {len(records)} records")
+
+    print("Transforming to new schema...")
+    transformed = []
+    total_refs = 0
+    total_citations = 0
+
+    for i, record in enumerate(records):
+        new_record = transform_record(record)
+        transformed.append(new_record)
+        total_refs += new_record['reference_count']
+        total_citations += new_record['citation_count']
+
+        if (i + 1) % 100000 == 0:
+            print(f"  Processed {i + 1} records...")
+
+    print(f"Transformation complete:")
+    print(f"  Total arXiv works: {len(transformed)}")
+    print(f"  Total references: {total_refs}")
+    print(f"  Unique citing works: {total_citations}")
+
+    # Define new schema
+    schema = pa.schema([
+        ('arxiv_doi', pa.string()),
+        ('arxiv_id', pa.string()),
+        ('reference_count', pa.int64()),
+        ('citation_count', pa.int64()),
+        ('cited_by', pa.list_(pa.struct([
+            ('doi', pa.string()),
+            ('matches', pa.list_(pa.struct([
+                ('raw_match', pa.string()),
+                ('reference', pa.string()),
+            ]))),
+        ]))),
+    ])
+
+    print(f"Writing {output_path}...")
+
+    # Convert to Arrow format
+    arxiv_dois = [r['arxiv_doi'] for r in transformed]
+    arxiv_ids = [r['arxiv_id'] for r in transformed]
+    reference_counts = [r['reference_count'] for r in transformed]
+    citation_counts = [r['citation_count'] for r in transformed]
+    cited_by_lists = [r['cited_by'] for r in transformed]
+
+    table = pa.table({
+        'arxiv_doi': arxiv_dois,
+        'arxiv_id': arxiv_ids,
+        'reference_count': reference_counts,
+        'citation_count': citation_counts,
+        'cited_by': cited_by_lists,
+    }, schema=schema)
+
+    pq.write_table(table, output_path, compression='snappy')
+
+    output_size = output_path.stat().st_size / (1024 * 1024)
+    print(f"Done! Output size: {output_size:.1f} MB")
+
+    # Verify by reading back
+    print("\nVerifying output...")
+    verify_table = pq.read_table(output_path)
+    print(f"  Rows: {verify_table.num_rows}")
+    print(f"  Schema: {verify_table.schema}")
+
+    # Show sample record
+    sample = verify_table.to_pylist()[0]
+    print(f"\nSample record:")
+    print(f"  arxiv_doi: {sample['arxiv_doi']}")
+    print(f"  arxiv_id: {sample['arxiv_id']}")
+    print(f"  reference_count: {sample['reference_count']}")
+    print(f"  citation_count: {sample['citation_count']}")
+    print(f"  First citing work: {sample['cited_by'][0]['doi']}")
+    print(f"  Matches in first citing work: {len(sample['cited_by'][0]['matches'])}")
+
+
+if __name__ == '__main__':
+    main()