enron-pii-detector

Project to extract PII in Enron emails dataset taken from Kaggle

Project Structure

/data: Contains the dataset. Only smaller subset of the dataset is uploaded to Github. The full dataset can be downloaded as per instructions in 00_get_data.ipynb
/data/labels/: Folder containing human generated labels considered the ground truth for this project.
/data/labels_llm/: Folder containing subfolders (basically a subfolder for each different approach or experiment) of LLM generated labels for evaluation.
/src/: Folder containing some helper functions and utility functions used in different places in this project.
labeling_app.py: A little labeling app built with Streamlit to label the dataset. It is this app that generates the ground truth labels in /data/labels/.
00_get_data.ipynb: A notebook to download the dataset from Kaggle, extract it, shuffle it, and do some random splits.
01_openai_prompt_engineering.ipynb: A notebook to generate labels for the /data/labels_llm/ folder using OpenAI's API and minimal prompt engineering. This is to serve as a baseline for the LLM approach(s).
02_openai_prompt_engineering_evaluation.ipynb: A notebook to evaluate the labels generated in 01_openai_prompt_engineering.ipynb using the ground truth labels in /data/labels/.