| license: cc-by-4.0 | |
| task_categories: | |
| - feature-extraction | |
| language: | |
| - en | |
| tags: | |
| - semantic-similarity | |
| - rare-words | |
| size_categories: | |
| - n<1K | |
| # Card-660 Dataset | |
| This is the **Card-660** (Cambridge Rare Word) dataset, converted to Hugging Face `datasets` format. | |
| ## Description | |
| Card-660 is a benchmark for evaluating semantic similarity models, specifically focusing on **rare and infrequent words**. It contains 660 word pairs annotated by experts. It was designed to solve the issues of low inter-annotator agreement found in other rare-word datasets (like RW). | |
| ## Columns | |
| - `word1`: First word of the pair | |
| - `word2`: Second word of the pair | |
| - `similarity`: Human-annotated similarity score (scale typically 0-4 or 0-10) | |
| ## Usage | |
| ```python | |
| from datasets import load_dataset | |
| dataset = load_dataset("Yuti/Card-660") | |