EMNLP 2023 Findings: Locally Differentially Private Document Generation Using Zero Shot Prompting Paper Link
This is the repository for the paper "Locally Differentially Private Document Generation Using Zero Shot Prompting," presented at EMNLP 2023 Findings. This paper introduces a method for generating differentially private text documents using zero-shot prompting techniques.
-
attacks: Directory containing Python scripts for attacks.
-
data: Directory for storing the data related to the project, including clean data and ChatGPT paraphrased data.
-
mechanisms: Directory for Python scripts related to mechanisms for document, sentence, and word-level tasks.
-
requirements.txt: File listing the required packages and dependencies for the project.
-
dp_prompt.ipynb: Demo notebook showing an experiment snapshot comparing DP-Prompt vs Madlib.
- Python 3.9+
- Dependencies listed in
requirements.txt - A suitable environment or container (e.g., virtual environment)
To set up the project on your local machine, follow these steps:
- git clone git@github.com:SaitejaUtpala/dp_prompt.git
- cd dp_prompt
- pip install -r requirements.txt
Demo notebook showing experiment snapshot comparing DP-Prompt vs Madlib clicking here.