Robust Heterogeneous Graph Neural Networks against Adversarial Attacks
Authors: Mengmei Zhang, Xiao Wang, Meiqi Zhu, Chuan Shi, Zhiqiang Zhang, Jun Zhou4363-4370
AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on different benchmark datasets for multiple HGNNs are conducted, where the considerable improvement of HGNNs under adversarial attacks will demonstrate the effectiveness and generalization ability of our defense framework. |
| Researcher Affiliation | Collaboration | Mengmei Zhang1, Xiao Wang1, Meiqi Zhu1, Chuan Shi1*, Zhiqiang Zhang2, Jun Zhou2 1Beijing University of Posts and Telecommunications 2Ant Group, Hangzhou, China |
| Pseudocode | Yes | Algorithm 1: Ro He: Robust heterogeneous HAN |
| Open Source Code | No | The paper does not provide an explicit statement about releasing source code or a link to a code repository. |
| Open Datasets | Yes | Ro He is evaluated on three widely used HG datasets: (1) ACM (Wang et al. 2019a) consists of {Paper (P), Author (A), Subject (S)}... (2) DBLP (Fu et al. 2020) consists of {Author (A), Paper (P), Term (T), Conference (C)}... (3) Aminer (Hu, Fang, and Shi 2019) consists of {Paper (P), Author (A), Reference (R)}... |
| Dataset Splits | No | The paper mentions evaluating performance with Micro-F1 metric over test set, but it does not explicitly provide details about training/validation/test splits (e.g., percentages or counts for validation data). |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for experiments. |
| Software Dependencies | No | The paper does not specify any software names with version numbers. |
| Experiment Setup | Yes | Given a target node, we limit adversarial edges with budget = {1, 2, 3} and edge types as P-A for ACM/DBLP and P-R for Aminer. We evaluate the performance with Micro-F1 metric over 500 target nodes, which are randomly sampled from the test set. Parameter Study: We analyze hyper-parameters T which is the number of neighbors to be kept in puriļ¬er of HAN-Ro He, under scenarios of Clean and Attack with = {1, 3}. |