Datasets:
Add dataset README
Browse files
README.md
ADDED
|
@@ -0,0 +1,49 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
task_categories:
|
| 4 |
+
- text-retrieval
|
| 5 |
+
language:
|
| 6 |
+
- en
|
| 7 |
+
tags:
|
| 8 |
+
- climate-fact-checking
|
| 9 |
+
- comp90042
|
| 10 |
+
- hard-negative-mining
|
| 11 |
+
---
|
| 12 |
+
|
| 13 |
+
# New-Methodology retrieval training data
|
| 14 |
+
|
| 15 |
+
Artifacts for the COMP90042 climate claim verification retrieval pipeline.
|
| 16 |
+
|
| 17 |
+
## Files
|
| 18 |
+
|
| 19 |
+
| File | Description |
|
| 20 |
+
|------|-------------|
|
| 21 |
+
| `reranker_train_pairs.jsonl` | CE finetune pairs (claim, passage, label) mined from train |
|
| 22 |
+
| `reranker_train_pairs.meta.json` | Mining metadata |
|
| 23 |
+
| `train_bge_top150_candidates.json` | Per-claim BGE-large top-150 candidate pools (train) |
|
| 24 |
+
|
| 25 |
+
## Mining config (from meta)
|
| 26 |
+
|
| 27 |
+
```json
|
| 28 |
+
{
|
| 29 |
+
"output": "/hpc2hdd/home/jyang577/joe/New-Methodology/data/reranker_train_pairs.jsonl",
|
| 30 |
+
"pool_cache": "/hpc2hdd/home/jyang577/joe/New-Methodology/data/train_bge_top150_candidates.json",
|
| 31 |
+
"dense_model": "BAAI/bge-large-en-v1.5",
|
| 32 |
+
"candidate_top_n": 150,
|
| 33 |
+
"hard_neg_min": 4,
|
| 34 |
+
"hard_neg_max": 8,
|
| 35 |
+
"bf16_encode": false,
|
| 36 |
+
"num_pairs": 13946,
|
| 37 |
+
"claims": 1228,
|
| 38 |
+
"positives": 4122,
|
| 39 |
+
"hard_negatives": 9824,
|
| 40 |
+
"claims_with_fewer_than_min_neg": 0,
|
| 41 |
+
"claims_with_no_gold": 0
|
| 42 |
+
}
|
| 43 |
+
```
|
| 44 |
+
|
| 45 |
+
## Regenerate locally
|
| 46 |
+
|
| 47 |
+
```bash
|
| 48 |
+
python scripts/01_mine_hard_negatives.py
|
| 49 |
+
```
|