Theoretically Guaranteed Bidirectional Data Rectification for Robust Sequential Recommendation

Authors: Yatong Sun, Bin Wang, Zhu Sun, Xiaochun Yang, Yan Wang

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on four real-world datasets verify the generality, effectiveness, and efficiency of our proposed Bir DRec.
Researcher Affiliation Collaboration Yatong Sun1,2, Xiaochun Yang1 , Zhu Sun3 , Bin Wang1, Yan Wang2 1School of Computer Science and Engineering, Northeastern University, China 2School of Computing, Macquarie University, Australia 3Center for Frontier AI Research, Institute of High Performance Computing, A*STAR, Singapore
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes Our code is available at: https://github.com/AlchemistYT/BirDRec.
Open Datasets Yes Datasets. We adopt four real-world datasets with varying domains, sizes, sparsity, and average sequence lengths shown in Table 1. Specifically, ML-1M (ML)[45] is a popular movie recommendation benchmark. Beauty (Be) [46] is the product review dataset collected from Amazon.com. Yelp (Ye) [10] is a business recommendation dataset released by Yelp.com. QK-Video (QK) [47] is a video recommendation dataset crawled from Tencent.com.
Dataset Splits Yes For each user, we preserve the last two interactions for validation and testing, while the rest are used for training.
Hardware Specification Yes All the experiments are conducted on an NVIDIA Quadro RTX 8000 GPU.
Software Dependencies No The paper mentions "Py Torch" but does not specify a version number.
Experiment Setup Yes For all methods, Xavier initializer [51] and Adam optimizer [52] are adopted; and the best hyper-parameter settings are empirically found based on the performance on the validation set. For Bir DRec, it is implemented by Py Torch with batch_size = 1024, d = 64, learning_rate = 0.01 for Yelp and 0.001 for other datasets, L = 5, β = 0.1, β = 0.1, K = 10, and ρ = 0.9. To ensure accurate rectification with reduced ϵ, Bir DRec is trained without rectification in the first 10 epochs.