Update README.md
Browse files
README.md
CHANGED
|
@@ -38,3 +38,55 @@ configs:
|
|
| 38 |
- split: validation
|
| 39 |
path: data/validation-*
|
| 40 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 38 |
- split: validation
|
| 39 |
path: data/validation-*
|
| 40 |
---
|
| 41 |
+
# Reference Parsing Finetuning Dataset
|
| 42 |
+
|
| 43 |
+
A fine-tuning dataset for bibliographic reference extraction and parsing, combining LinkedBook, CEX, and EXCITE datasets into conversation-style examples for LLM training.
|
| 44 |
+
|
| 45 |
+
## Dataset Description
|
| 46 |
+
|
| 47 |
+
This dataset teaches models to extract and parse bibliographic references from text into structured JSON format. Examples follow a conversational format with system/user/assistant messages, using various prompt variants for diversity.
|
| 48 |
+
|
| 49 |
+
**Data Sources:**
|
| 50 |
+
- **LinkedBooks**: Multi-language reference strings with structured metadata
|
| 51 |
+
- **CEX**: English academic papers with TEI XML parsed references
|
| 52 |
+
- **EXCITE**: Multi-language academic papers with parsed references
|
| 53 |
+
|
| 54 |
+
## Data Fields
|
| 55 |
+
|
| 56 |
+
| Field | Type | Description |
|
| 57 |
+
|-------|------|-------------|
|
| 58 |
+
| `messages` | list | Conversation messages with `role` (system/user/assistant) and `content` |
|
| 59 |
+
| `source` | string | Data source: `linkedbook`, `cex`, or `excite` |
|
| 60 |
+
| `split` | string | Dataset split: `train` or `valid` |
|
| 61 |
+
| `mode` | string | Example type: `single` (1 reference) or `group` (multiple references) |
|
| 62 |
+
| `language` | string | Language code (e.g., `en`, `de`, `fr`) |
|
| 63 |
+
| `ref_count` | int | Number of references in the example |
|
| 64 |
+
| `file_id` | string\|null | Source document ID (for CEX/EXCITE) |
|
| 65 |
+
| `category` | string\|null | Document category (for CEX) |
|
| 66 |
+
|
| 67 |
+
## Splits
|
| 68 |
+
|
| 69 |
+
| Split | Examples | Description |
|
| 70 |
+
|-------|----------|-------------|
|
| 71 |
+
| `train` | ~1,708 | Main training data |
|
| 72 |
+
| `valid` | ~115 | Validation set |
|
| 73 |
+
|
| 74 |
+
**Distribution:**
|
| 75 |
+
- ~70% single-reference examples, ~30% multi-reference groups
|
| 76 |
+
- ~10-15% LinkedBook, ~30-35% CEX, ~50-55% EXCITE
|
| 77 |
+
|
| 78 |
+
## Data Creation and Processing
|
| 79 |
+
|
| 80 |
+
|
| 81 |
+
1. **Data Loading**: Loads references from LinkedBooks (Training and Validation JSONL), CEX (JSON + TEI XML), and EXCITE (JSON + XML)
|
| 82 |
+
2. **Validation**: Filters invalid references (missing titles/authors, unparsed authors, mismatched counts)
|
| 83 |
+
3. **Sampling**: Stratified sampling by category/class (30% train rate for CEX/EXCITE)
|
| 84 |
+
4. **Grouping**: Groups references into batches (3-20 refs per group with weighted probabilities)
|
| 85 |
+
5. **Prompt Variants**: Applies 5 prompt variants with weighted distribution (40% detailed, 25% minimal, 25% task-based, 5% ultra-minimal, 5% no prompt)
|
| 86 |
+
6. **Format Conversion**: Converts to conversation-style format with structured JSON output
|
| 87 |
+
|
| 88 |
+
|
| 89 |
+
|
| 90 |
+
## Credits
|
| 91 |
+
|
| 92 |
+
The dataset is being developed by [Yurui Zhu](https://github.com/RuiaRui) ([Odoma](https://github.com/odoma-ch)). This work is carried out in the context of the EU-funded [GRAPHIA project](https://graphia-ssh.eu/) (grant ID: [101188018](https://cordis.europa.eu/project/id/101188018)).
|