|
|
--- |
|
|
dataset_info: |
|
|
features: |
|
|
- name: Judgement-Summary Pairs |
|
|
dtype: string |
|
|
splits: |
|
|
- name: train |
|
|
num_bytes: 210389804 |
|
|
num_examples: 7030 |
|
|
- name: test |
|
|
num_bytes: 3265635 |
|
|
num_examples: 100 |
|
|
download_size: 98328073 |
|
|
dataset_size: 213655439 |
|
|
configs: |
|
|
- config_name: default |
|
|
data_files: |
|
|
- split: train |
|
|
path: data/train-* |
|
|
- split: test |
|
|
path: data/test-* |
|
|
--- |
|
|
|
|
|
# Reformatted Legal Text Summarization Dataset |
|
|
|
|
|
## Overview |
|
|
|
|
|
This dataset is a restructured version of a Legal Text Summarization dataset, specifically filtered to include only entries from the `IN-Abs` dataset. The dataset consists of pairs of legal judgements and their corresponding summaries. Each judgement-summary pair is formatted into an **Instruction-Prompt style format** suitable for training language models for tasks like Summarization and Legal Text Analysis. |
|
|
|
|
|
## Dataset Structure |
|
|
|
|
|
- **Column Name:** `Judgement-Summary Pairs` |
|
|
- **Format:** The dataset contains one column named "Judgement-Summary Pairs", which holds the formatted instruction-prompt pairs. Each entry is a structured pair of a legal judgement and its corresponding summary. |
|
|
|
|
|
### Example of an Entry: |
|
|
|
|
|
```plaintext |
|
|
<s>[INST] Summarize the following judgement: [Judgement Text] [/INST] [Summary Text] </s> |
|
|
``` |
|
|
Where: |
|
|
[Judgement Text] is the original legal judgement text. |
|
|
[Summary Text] is the summary of the judgement. |
|
|
|
|
|
### Sample Entry: |
|
|
|
|
|
```plaintext |
|
|
<s>[INST] Summarize the following judgement: The court finds that the defendant was not responsible for the damages caused as there was insufficient evidence to support the claims made by the plaintiff. [/INST] The court ruled in favor of the defendant due to lack of evidence. </s> |
|
|
``` |
|
|
|
|
|
## Data Preparation Process: |
|
|
The dataset was prepared using the following steps: |
|
|
- Filtering: The dataset was filtered to include only entries from the IN-Abs dataset. |
|
|
- Transformation: Each judgement and its corresponding summary were formatted into instruction-prompt pairs using the following template: |
|
|
- ```plaintext |
|
|
<s>[INST] Summarize the following judgement: {judgement_text} [/INST] {summary_text} </s> |
|
|
``` |
|
|
- Column Removal: All columns except the newly created "Judgement-Summary Pairs" were removed to streamline the dataset for usage. |
|
|
|
|
|
## Usage |
|
|
The reformatted dataset is suitable for training language models on legal text summarization tasks. Here’s how you can use it: |
|
|
|
|
|
- **Load the Dataset**: Use a data loading library like datasets or pandas to read the dataset. |
|
|
- **Training a Model**: You can train a model using the instruction-prompt pairs provided in the dataset. Each pair serves as an input-output example for supervised learning, particularly in the fields of natural language processing (NLP) and legal AI. |
|
|
- **Fine-tuning Existing Models**: Fine-tune pre-trained models such as GPT, BERT, or LLaMA on this dataset to adapt them for legal text summarization tasks. |
|
|
|
|
|
### License |
|
|
The dataset is made available for educational and research purposes. Please ensure that you have the appropriate permissions and rights to use the original data before utilizing this reformatted version. |
|
|
|
|
|
### Acknowledgements |
|
|
The dataset preparation and transformation were inspired by the need to improve legal text summarization and the creation of structured datasets for NLP tasks. |
|
|
Special thanks to the original data providers and contributors to legal text summarization research |
|
|
|
|
|
### Contact |
|
|
For questions or further information, please contact **Ajay Mukund S** at *[email protected]*. |
