WIP Research

encRec

Trie-routed recommendation under encrypted features for privacy-preserving ML

// DESCRIPTION

Research on recommendation systems that can operate on encrypted user features using Trie-based routing.

Problem

How to make recommendations without exposing raw user data?

Approach

  • Statistical Trie: Organize encrypted features hierarchically
  • LLM-assisted Routing: Guide queries through Trie structure
  • Privacy Guarantees: Features remain encrypted throughout

Datasets

  • Criteo (click prediction)
  • Amazon (product recommendation)
  • Avazu (ad CTR prediction)

// HIGHLIGHTS

  • Privacy-preserving recommendation
  • Novel Trie-based approach
  • Multi-dataset evaluation

TECH_STACK

Python PyTorch DeepFM DLRM Homomorphic Encryption

PROJECT_INFO

started: 2024-10-01
status: WIP
type: Research