Update README.md
Browse files
README.md
CHANGED
|
@@ -50,16 +50,13 @@ the full field-level specification. Key fields:
|
|
| 50 |
- `websci_meta.sci_quality_score`: 0–5 scientific value score from the
|
| 51 |
rubric-based quality classifier.
|
| 52 |
|
| 53 |
-
## Curation
|
|
|
|
| 54 |
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
1. Heuristic filtering with discipline-aware differential rules.
|
| 60 |
-
2. Two-tier content-driven discipline routing.
|
| 61 |
-
3. Rubric-based quality scoring with asymmetric penalties.
|
| 62 |
-
4. Intra-dump fuzzy + intra-disciplinary exact deduplication.
|
| 63 |
|
| 64 |
## PII Handling
|
| 65 |
|
|
|
|
| 50 |
- `websci_meta.sci_quality_score`: 0–5 scientific value score from the
|
| 51 |
rubric-based quality classifier.
|
| 52 |
|
| 53 |
+
## Data Curation Pipeline
|
| 54 |
+
The HainaWeb-Sci corpus is constructed from DCLM-Pool and DCLM-Pool-aligned Common Crawl data (2013–March 2026). We process the raw data through a four-stage data-centric curation pipeline tailored to scientific web data:
|
| 55 |
|
| 56 |
+
- Stage 1 (Data Preparation) transforms raw web data into a consistent textual foundation through WARC parsing, text extraction, URL filtering, and language identification.
|
| 57 |
+
- Stage 2 (Quality Filtering) focuses on science-aware content refinement by combining rule-based and model-based filtering. Unlike general-purpose pipelines, both heuristic rules and learned quality models are customized to better preserve scientifically meaningful content, particularly symbolic expressions and structured reasoning.
|
| 58 |
+
- Stage 3 (Deduplication) removes redundant content at scale to improve corpus diversity and training efficiency.
|
| 59 |
+
- Stage 4 (Discipline-Specific Curation) introduces fine-grained discipline classification and scientific value scoring, enabling structured organization of scientific knowledge across 14 disciplines and providing quantitative signals for data selection.
|
|
|
|
|
|
|
|
|
|
|
|
|
| 60 |
|
| 61 |
## PII Handling
|
| 62 |
|