metadata
task_categories:
- token-classification
language:
- it
size_categories:
- 100K<n<1M
Dataset Overview
The dataset provided here is a comprehensive collection supporting the model's training and evaluation. It is particularly suited for those looking to work on or improve Italian NER systems.
Key Features
- Total Texts: The dataset contains over 110,000 annotated texts, providing a rich resource for training and testing NER models.
- Unique Labels: There are a total of 97,351 different labels within the dataset. These labels cover a broad spectrum of named entities, making the dataset invaluable for developing models capable of identifying a wide variety of entities.
- Annotation Quality: Annotations have been meticulously verified to ensure high accuracy and reliability for model training and benchmarking.