Uplift Modeling for Target User Attacks on Recommender Systems

This repository contains our implementations for UBA(an Uplift-guided Budget Allocation Framework) and various shilling attack methods including Leg-UP, AIA, WGAN, Random Attack, Average Attack, Segment Attack and Bandwagon Attack.

Enviroment

conda env create -f environment.yml
conda activate uba

Data

You can download ML-1M, Yelp, Amazon

Usage

Attack

1. Attack LightGCN model with baseline attacker Leg-UP

python -u main.py --data_set ml1m --model lgn --attacker_list  AUSHplus --attack_num 300 --oneitem 3116  --allseed 2023 --way 1

2. Attack mf model with UBA attacker AIA

python -u main.py --data_set ml1m --model mf --attacker_list  AIA --attack_num 300 --oneitem 3116  --allseed 2023 --way 3

Dynamic Programming Algorith

python DPA.py

For more example and parameters, please reference the example.sh .

Continue updating...