Datasets:

FedPII
/

FedPII-dataset

license: cc-by-nc-4.0
task_categories:
  - text-classification
  - question-answering
  - summarization
  - text-retrieval
  - text-generation
language:
  - zh
tags:
  - legal
  - privacy
  - federated-learning

FedPII Dataset

This repository contains the dataset for the paper Simple Yet Effective: Extracting Private Data Across Clients in Federated Fine-Tuning of Large Language Models.

FedPII is a Chinese legal-domain dataset with fine-grained PII (Personally Identifiable Information) annotations consistent with CPIS, GDPR, and CCPA standards. It is designed to evaluate privacy risks and cross-client data leakage in Federated Large Language Models (FedLLMs).

GitHub Repository: SMILELab-FL/FedPII

Dataset Description

The dataset consists of 16,194 samples across various legal-related tasks. It includes fine-grained PII annotations (such as names, addresses, and birthdays) used to assess how easily private information can be recovered during federated learning.

Tasks and Statistics

Task	Description	Split	Samples	With PII
`leg_case_cls`	Legal Case Classification	Train/Test	4,028	1,019
`jud_sum`	Judicial Summarization	Train/Test	2,651	2,650
`sim_case_match`	Similar Case Matching	Train/Test	3,773	3,771
`jud_read_compre`	Judicial Reading Comprehension	Train/Test	3,498	3,493
`exam`	Legal Exam	Train/Test	2,244	700

Total Samples: 16,194 (14,575 Train / 1,619 Test).

Citation

@inproceedings{hu-etal-2025-simple,
    title = "Simple Yet Effective: Extracting Private Data Across Clients in Federated Fine-Tuning of Large Language Models",
    author = "Hu, Yingqi  and
      Zhang, Zhuo  and
      Zhang, Jingyuan  and
      Wang, Jinghua  and
      Wang, Qifan  and
      Qu, Lizhen  and
      Xu, Zenglin",
    booktitle = "Proceedings of the 14th International Joint Conference on Natural Language Processing and the 4th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics",
    month = dec,
    year = "2025",
    address = "Mumbai, India",
    publisher = "The Asian Federation of Natural Language Processing and The Association for Computational Linguistics",
    url = "https://aclanthology.org/2025.findings-ijcnlp.113/",
    pages = "1808--1827",
    ISBN = "979-8-89176-303-6",
}

License

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.