Sim4Rec: Data-Free Model Extraction Attack on Sequential Recommendation

Authors: Yihao Wang, Jiajie Su, Chaochao Chen, Meng Han, Chi Zhang, Jun Wang

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments demonstrate the advancement of Sim4Rec. (3) Extensive experiments on three benchmark datasets demonstrate that Sim4Rec outperforms existing methods in extraction performance and improves downstream task outcomes. In this section, we conduct experiments to address the following research questions. Datasets and models. We evaluate our method on three benchmark datasets, i.e., Movie Lens-1M (ML-1M), Steam, and Beauty. Evaluation protocols. Following (Yue et al. 2021b), We evaluate our method from three metrics.
Researcher Affiliation Collaboration Yihao Wang1, Jiajie Su1, Chaochao Chen1, Meng Han1, Chi Zhang2, Jun Wang3* 1Zhejiang University 2Independent Researcher 3OPPO Research Institute EMAIL, EMAIL, EMAIL
Pseudocode No The paper describes methods and algorithms using mathematical formulations and textual explanations, such as in Section 4 (Method) and its subsections (Controllable Sequence Generation, Reinforced Adversarial Distillation), but it does not contain a clearly labeled pseudocode block or algorithm figure.
Open Source Code No The paper does not contain any explicit statements about releasing source code, nor does it provide a link to a code repository. The text only discusses the methodology and experimental results without mentioning code availability.
Open Datasets Yes We evaluate our method on three benchmark datasets, i.e., Movie Lens-1M (ML-1M), Steam, and Beauty. Details of datasets are shown in Table 1. Table 1: Statistics of datasets. ML-1M 6,040 3,416 1.0M 163.5 4.79% Steam 334,730 13,047 3.7M 10.6 0.08% Beauty 40,226 54,542 0.4M 8.8 0.02%
Dataset Splits Yes We reserve the last two items in each sequence for validation and testing, and the remaining items used for training.
Hardware Specification No The paper describes implementation details such as the model architecture, optimizer, learning rate, and batch size, but it does not specify any hardware details like GPU/CPU models or memory specifications used for running the experiments.
Software Dependencies No The paper mentions using Gated Recurrent Unit (GRU) as the architecture of the sequential generator and Adam optimizer. However, it does not provide specific version numbers for any programming languages, libraries, or frameworks (e.g., Python version, PyTorch/TensorFlow version).
Experiment Setup Yes We utilize Gated Recurrent Unit (GRU) as the architecture of the sequential generator. Adam optimizer with learning rate γ = 0.01, weight decay η = 0.01, and batch size b = 128 are adopted. We set the allowed sequence lengths of ML-1M, Steam, and Beauty to {200, 50, 50}, the hyper-parameters to Emle = 30 and Mbank = 100.