Single-Node Trigger Backdoor Attacks in Graph-Based Recommendation Systems

Authors: Runze Li, Di Jin, Xiaobao Wang, Dongxiao He, Bingdao Feng, Zhen Wang

IJCAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results show that the exposure of the target items reaches no less than 50% in 99% of the target users, while the impact on the recommendation system s performance is controlled within approximately 5%. Additionally, Section 4 is titled "Experiments" and discusses evaluation on datasets with various metrics.
Researcher Affiliation Academia The authors are affiliated with "Tianjin University", "Guangdong Laboratory of Artificial Intelligence and Digital Economy(SZ)", and "Northwestern Polytechnical University". All email addresses end in '.edu.cn', indicating academic affiliations.
Pseudocode No The paper describes the proposed method, including the trigger generation, GNN-based recommendation, and joint optimization function, using prose and mathematical equations. There are no clearly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not contain any explicit statements about releasing source code for the described methodology, nor does it provide links to a code repository.
Open Datasets Yes The paper states: "We conduct experiments on four real-world datasets, namely Gowalla, Amazon, Yelp and Movie Lens." and provides URLs for each: "Gowalla1: http://snap.stanford.edu/data/loc-gowalla.html Amazon2: https://snap.stanford.edu/data/amazon/ Yelp3: https://www.kaggle.com/yelp-dataset/yelp-dataset Movie Lens4: https://www.kaggle.com/datasets/movielens-100k-dataset"
Dataset Splits Yes In the "Experimental Settings" section, the paper states: "To evaluate the recommendation performance, we split each dataset into two parts: 80% for training and 20% for testing."
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types, or memory amounts) used for running the experiments.
Software Dependencies No The paper mentions using Light GCN as the surrogate model but does not provide specific software names with version numbers for programming languages, libraries, or solvers used in the experimental setup.
Experiment Setup Yes In the "Experimental Settings" section, the paper specifies: "The training epoch, node embedding size, and learning rate are set to 1000, 64, and 0.001, respectively." It also states: "The number of target items is set to 20." and mentions hyperparameters α, β, and γ for the joint optimization function.