amityahav / ppcf

Mediated Secure Multi-Party Protocols for Collaborative Filtering

Privacy Preserving Collaborative Filtering

Mediated Secure Multi-Party Protocols for Collaborative Filtering

Implementation of Erez Shmueli and Tamir Tassa's CF algorithm in a privacy preserving manner as written in their paper:

Erez Shmueli, Tamir Tassa: Mediated Secure Multi-Party Protocols for Collaborative Filtering. (2020)

Setup Guide:

1. git clone the repository
2. cd to the repo directory
3. run docker-compose up (Make sure Docker is up and running)
4. Vendors service is accessible via http://127.0.0.1:5003, Mediator service is accessible via http://127.0.0.1:5004

Vendors Service API endpoints:

• /start - Starting Offline phase of the Algorithm for K vendors (Model Construction)
• /predict - For a given vendor id, user id, and item id, returns a prediction for this user rating of the item
• /most_recommended - For a given vendor id and user id, returns the h-most recommended items for this specific user
• /error - Calculating Mean Absolute Error over the u1.test test set (located in vendors/app/data/raw_data)

Model was already computed locally on the MovieLen-100k dataset ,80/20 spllit, and used in both services.

• for a fresh computation of the Model this 3 lines inside the Vendors Dockerfile need to be commented out :

COPY app/data/similarity_matrix.npy ppcf/mediator/app/data
COPY app/data/encrypted_mask.npy ppcf/mediator/app/data
COPY app/data/encrypted_user_item_matrix.npy ppcf/mediator/app/data