PANORAMA Plus: Extra informationProfile-based Assemblage for Naturalistic Online Representation and Attribute Memorization Analysis
A Synthetic PII-Laced Dataset for Studying Sensitive Data Memorization in LLMs
Dataset Summary
PANORAMA is a large-scale, fully synthetic dataset comprising 384,789 samples derived from 9,674 internally consistent synthetic profiles. The dataset closely emulates the distribution, variety, and context of personally identifiable information (PII) and other sensitive data as it naturally occurs in online environments. Each sample is generated using a combination of prompt engineering and reasoning LLMs, spanning diverse content types such as wiki-style articles, social media posts, forum discussions, online reviews, comments, and marketplace listings.
PANORAMA is designed for systematically studying the memorization of sensitive information by large language models (LLMs) and for evaluating privacy-preserving strategies.
PANORAMA Plus contains more details on PANORAMA dataset released by us, use the unique ID field to connect back with PANORAMA dataset.
Attribution
Created by Sriram Selvam and Anneswa Ghosh.
If you use this dataset, please cite:
Citation
@article{selvam2025panorama,
title={PANORAMA: A synthetic PII-laced dataset for studying sensitive data memorization in LLMs},
author={Selvam, Sriram and Ghosh, Anneswa},
journal={arXiv preprint arXiv:2505.12238},
year={2025}
}
License
Creative Commons Attribution 4.0 International (CC BY 4.0)
- Downloads last month
- 103