| --- |
| pretty_name: LexBench-SWDE-800 |
| license: other |
| license_name: microsoft-reciprocal-license-ms-rl |
| license_link: https://huggingface.co/datasets/Lexmount/LexBench-SWDE-800/blob/main/LICENSE.txt |
| task_categories: |
| - other |
| size_categories: |
| - n<1K |
| tags: |
| - swde |
| - structured-data-extraction |
| - benchmark |
| configs: |
| - config_name: default |
| data_files: |
| - split: test |
| path: inputs.jsonl |
| --- |
| |
| # SWDE-800: fixed evaluation input subset of SWDE |
|
|
| Short name: **LB-SWDE**. |
|
|
| This distribution contains 800 archived HTML pages from **SWDE (Structured Web Data Extraction)**, selected for a Lexmount API extraction evaluation. SWDE was created by its original authors. Lexmount selected and packaged these inputs; it does not claim authorship of the source benchmark. |
|
|
| The original collection has 124,291 pages from 80 websites in eight domains. This subset contains ten pages from every website: 100 pages each for automobiles, books, cameras, jobs, movies, NBA players, restaurants and universities. It is a predeclared evaluation subset, not an official SWDE split or the complete benchmark. |
|
|
| ## Source and selection |
|
|
| The original project was hosted at [CodePlex](https://archive.codeplex.com/?p=swde). These files come from the [SWDE source mirror at the pinned revision](https://github.com/woailaosang/swde/tree/e9b60dbbcb899d70e813e8b899b84e3540d540da). The original description is preserved in `UPSTREAM_README.txt`. |
|
|
| For each website, pages were ordered by the SHA-256 digest of `seed + "\0" + original_relative_path`, using seed `swde-native-api-first-round-20261104`; the first ten pages were selected before API evaluation. `selection.json` records the source archive digest and selection rule. `provenance.json` maps each case to its original path and input digest. |
|
|
| ## Files and use |
|
|
| `inputs.jsonl` contains one row per page, with its ID, domain, website, original URL, original path, SHA-256 digest and HTML string. Encoding the HTML string as UTF-8 recovers the original bytes. The archived `<base>` element and external resource references are preserved. These pages describe historical website content; current URLs may have changed or disappeared. |
|
|
| The original SWDE task extracts 32 attribute types across the eight domains. Its annotations were created with handcrafted regular expressions at DOM text-node level; one field can contain multiple values. This input-only distribution does not include reference answers, provider responses or evaluation scores. Consult the original benchmark for its annotation files and task conventions. |
|
|
| ## License and attribution |
|
|
| `LICENSE.txt` preserves the original CodePlex project's Microsoft Reciprocal License (Ms-RL). That project license describes accompanying software; captured third-party website content remains subject to its publishers' rights. The repository metadata records the original license and this distinction rather than assigning a new blanket license to the pages. No additional MIT, Apache or Creative Commons license is granted by this redistribution. |
|
|
| Please cite SWDE and retain its original source attribution when using these files. Credit Lexmount only for the 800-page selection and this packaging. |
|
|
| ## Public HTML snapshots |
|
|
| [snapshots.json](snapshots.json) binds every page ID to its SHA-256 and an immutable public deployment. Native extraction APIs receive `fixture_base + '/' + id`; the original HTML files remain in this repository. |
|
|
| Original files are authoritative. Publication verifies the declared dataset version and complete input inventory before adding this binding. Changed HTML requires a new deployment and a new HF commit; previous files and deployment URLs remain available. Pin the full HF commit for evaluation and verify page hashes before and after calls. A mismatch stops collection rather than silently replacing the input. The navigation index does not host dataset HTML. |
|
|
