File size: 1,165 Bytes
7cd170a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
---
license: mit
task_categories:
  - text-retrieval
language:
  - en
tags:
  - climate-fact-checking
  - comp90042
  - hard-negative-mining
---

# New-Methodology retrieval training data

Artifacts for the COMP90042 climate claim verification retrieval pipeline.

## Files

| File | Description |
|------|-------------|
| `reranker_train_pairs.jsonl` | CE finetune pairs (claim, passage, label) mined from train |
| `reranker_train_pairs.meta.json` | Mining metadata |
| `train_bge_top150_candidates.json` | Per-claim BGE-large top-150 candidate pools (train) |

## Mining config (from meta)

```json
{
  "output": "/hpc2hdd/home/jyang577/joe/New-Methodology/data/reranker_train_pairs.jsonl",
  "pool_cache": "/hpc2hdd/home/jyang577/joe/New-Methodology/data/train_bge_top150_candidates.json",
  "dense_model": "BAAI/bge-large-en-v1.5",
  "candidate_top_n": 150,
  "hard_neg_min": 4,
  "hard_neg_max": 8,
  "bf16_encode": false,
  "num_pairs": 13946,
  "claims": 1228,
  "positives": 4122,
  "hard_negatives": 9824,
  "claims_with_fewer_than_min_neg": 0,
  "claims_with_no_gold": 0
}
```

## Regenerate locally

```bash
python scripts/01_mine_hard_negatives.py
```