|
--- |
|
task_categories: |
|
- text-generation |
|
language: |
|
- zh |
|
--- |
|
# Dataset Card for AgentCourt |
|
|
|
## Dataset Description |
|
- **Repository:** [https://huggingface.co/datasets/youzi517/AgentCourt](https://huggingface.co/datasets/youzi517/AgentCourt) |
|
- **Paper:** [AgentCourt: Simulating Court with Adversarial Evolvable Lawyer Agents](https://arxiv.org/abs/2408.08089) |
|
- **GitHub:** [https://github.com/relic-yuexi/AgentCourt](https://github.com/relic-yuexi/AgentCourt) |
|
|
|
### Dataset Summary |
|
AgentCourt is a dataset created for simulating court scenarios with adversarial evolvable lawyer agents. It contains 550 cases, each consisting of statements from both the plaintiff and the defendant. This dataset is designed to support research in legal AI, particularly in the areas of adversarial reasoning and court simulation. For more detailed information, code, and resources related to this project, please visit our [GitHub repository](https://github.com/relic-yuexi/AgentCourt). |
|
|
|
### Supported Tasks and Leaderboards |
|
- `text-generation`: The dataset can be used to fine-tune language models for generating legal arguments or responses in a court setting. |
|
|
|
### Languages |
|
The dataset is primarily in Chinese, focusing on legal language and terminology used in court settings. |
|
|
|
## Dataset Structure |
|
|
|
### Data Instances |
|
Here's an example of a data instance in the dataset: |
|
```json |
|
{ |
|
"caseId": "xxx", |
|
"plaintiff_statement": "xxx", |
|
"defendant_statement": "xxx" |
|
} |
|
``` |
|
|
|
### Data Fields |
|
- `caseId`: A string identifier for each unique case. |
|
- `plaintiff_statement`: A string containing the statement or argument presented by the plaintiff. |
|
- `defendant_statement`: A string containing the statement or argument presented by the defendant. |
|
|
|
### Data Splits |
|
The dataset contains 550 cases in total. Users can divide this into training and testing sets according to their specific needs. The paper does not specify a predetermined split. |
|
|
|
## Dataset Creation |
|
|
|
### Curation Rationale |
|
This dataset was created to support research in simulating court scenarios with AI agents, particularly focusing on adversarial interactions between lawyer agents. The goal is to provide a resource for developing and testing AI models capable of understanding and generating legal arguments. |
|
|
|
## Considerations for Using the Data |
|
|
|
### Social Impact of Dataset |
|
This dataset has the potential to contribute to the development of AI systems that can assist in legal processes, potentially improving access to legal services and enhancing the efficiency of court proceedings. However, it's important to note that AI systems trained on this data should not be seen as replacements for human legal professionals, but rather as tools to augment their capabilities. |
|
|
|
### Discussion of Biases |
|
Users of this dataset should be aware that it may contain biases inherent in the legal system from which the cases are drawn. These could include biases related to language use, socioeconomic factors, or specific legal jurisdictions. Researchers and practitioners should carefully consider these potential biases when developing models based on this data. |
|
|
|
## Additional Information |
|
|
|
### Dataset Curators |
|
This dataset was curated by Guhong Chen, Liyang Fan, Zihan Gong, Nan Xie, Zixuan Li, Ziqiang Liu, Chengming Li, Qiang Qu, Shiwen Ni, and Min Yang. |
|
|
|
### Citation Information |
|
If you use this dataset in your research, please cite: |
|
``` |
|
@misc{chen2024agentcourtsimulatingcourtadversarial, |
|
title={AgentCourt: Simulating Court with Adversarial Evolvable Lawyer Agents}, |
|
author={Guhong Chen and Liyang Fan and Zihan Gong and Nan Xie and Zixuan Li and Ziqiang Liu and Chengming Li and Qiang Qu and Shiwen Ni and Min Yang}, |
|
year={2024}, |
|
eprint={2408.08089}, |
|
archivePrefix={arXiv}, |
|
primaryClass={cs.CL}, |
|
url={https://arxiv.org/abs/2408.08089}, |
|
} |
|
``` |
|
|
|
### Contributions |
|
Thanks to the authors of the "AgentCourt: Simulating Court with Adversarial Evolvable Lawyer Agents" paper for creating and sharing this dataset. For more information, code, and resources, please visit our [GitHub repository](https://github.com/relic-yuexi/AgentCourt). |
