Paraguay Legislation

The Paraguay Legislation dataset is a comprehensive collection of legal documents sourced from the legislative framework of Paraguay. The dataset contains legal documents sourced from the legislative framework of Paraguay, including resolutions, decrees, laws, and other kinds of legislative texts.

This dataset has been curated as a valuable resource for Natural Language Processing (NLP) tasks. The data is designed for research focused on text classification tasks. The classification process is divided into two objectives:

Binary classification: 0 - no cost and 1 - cost (legislation has costs for the society)
Multi-classification: classify the document into several hierarchical categories of costs.

For more information about multi-classification definitions, please check this link: <todo: link to>.

Subsets

The dataset contains various subsets, each representing different data quality and preparation stages. Within these subsets, you'll encounter multiple versions of the same data, with variations primarily reflecting differences in data quality, metadata columns, and preprocessing tasks applied to change the data.

The subsets are the following:

1. Raw: Data extracted from the sources files (URls, PDFs and Word files) without any transformation or sentence splitter. It can be helpful because you can access the raw data extracted from the seeds (PDFs and Word files) and apply other preprocessing tasks from this point to prepare the data without returning to extract texts from source files.

2. Sentences: Normalized data split by sentence, mainly treating issues of text extracted from PDF. This stage also adds metadata about the sentence, for example: if it is a title or not.

3. Sentence Unlabeled: Unlabeled corpora of Paraguay legislation. This data is prepared to be labeled by the experts. Each instance of the dataset represents a specific text passage, split by its original formatting extracted from raw text (from original documents).

4. Sentence labeled (Ground Truth): The labeled data is the ground truth data used to train the models. This data is annotated by legal experts indicating the existence of administrative costs (and other types) in the legislation. Each instance of the dataset represents a specific text passage.

This dataset has the following data splits:

Training Set: This portion of the data is used to train and fine-tune machine learning models.
Test Set: The test set is reserved for assessing the model's accuracy, generalization, and effectiveness. It remains unseen during training and helps gauge how well the model performs on new, unseen data.

Together, these labeled data subsets provide a crucial reference point for building and evaluating models, ensuring they can make informed predictions and classifications with high accuracy and reliability.