| # Sample Data Description |
| This directory contains **five example instances** sampled from the full set of Facet-aware Training Units (FTUs) constructed in the paper to illustrate the data format. |
| ## File Overview |
| The sample dataset consists of three JSON files: |
| - `selected_triplets.json` |
| - `selected_questions.json` |
| - `selected_metadata.json` |
| These three files are **aligned by paper_id** and should be used together as FTUs. |
| ## File Relationships |
| - `selected_triplets.json` |
| Defines **facet-aware doc-doc supervision** in the form of query–positive–negative triplets. |
| For each query paper, positives and negatives are specified separately for each facet, and all referenced paper_ids correspond to entries in `selected_metadata.json`. |
| - `selected_questions.json` |
| Contains **facet-specific questions** (background / method / result) constructed for each query paper. |
| Each entry consists of three questions and is linked to a query paper in `selected_triplets.json` via `paper_id`. |
| - `selected_metadata.json` |
| Contains paper-level metadata (title, abstract) and sentence-level facet annotations (`background`, `method`, `result`), identified by paper_ids. |
| ## Summary |
| Together, these files demonstrate how the full dataset supports **facet-aware scientific retrieval**, including: |
| - document content (`selected_metadata.json`), |
| - facet-oriented queries (`selected_questions.json`), and |
| - facet-specific relevance supervision (`selected_triplets.json`). |
