Self-Supervised Hypergraph Convolutional Networks for Session-based Recommendation

Authors: Xin Xia, Hongzhi Yin, Junliang Yu, Qinyong Wang, Lizhen Cui, Xiangliang Zhang4503-4511

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on three benchmark datasets demonstrate the superiority of our model over the SOTA methods, and the results validate the effectiveness of hypergraph modeling and self-supervised task.
Researcher Affiliation Academia 1The University of Queensland, 2Shandong University, 3King Abdullah University of Science and Technology {x.xia, h.yin1, jl.yu, qinyong.wang}@uq.edu.au, clz@sdu.edu.cn, xiangliang.zhang@kaust.edu.sa
Pseudocode No The paper includes Figure 1 illustrating the model pipeline but does not provide pseudocode or algorithm blocks.
Open Source Code Yes The implementation of our model is available via https://github.com/xiaxin1998/DHCN.
Open Datasets Yes We evaluate our model on two real-world benchmark datasets: Tmall2, Nowplaying3 and Diginetica4. Tmall dataset comes from IJCAI-15 competition, which contains anonymized user s shopping logs on Tmall online shopping platform. Nowplaying dataset describes the music listening behavior of users. For both datasets, we follow (Wu et al. 2019b; Li et al. 2017) to remove all sessions containing only one item and also remove items appearing less than five times. To evaluate our model, we split both datasets into training/test sets, following the settings in (Wu et al. 2019b; Li et al. 2017; Wang et al. 2020c). Then, we augment and label the dataset by using a sequence splitting method, which generates multiple labeled sequences with the corresponding labels ([is,1], is,2), ([is,1, is,2], is,3), ..., ([is,1, is,2, ..., is,m 1], is,m) for every session s = [is,1, is,2, is,3, ..., is,m]. Note that the label of each sequence is the last click item in it. The statistics of the datasets are presented in Table 1.
Dataset Splits No The paper mentions splitting data into "training/test sets" and references mini-batches during training, but does not explicitly state details about a distinct validation set split or its use.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU/GPU models, memory) used for running the experiments.
Software Dependencies No The paper does not specify any software dependencies with version numbers (e.g., Python version, library versions like PyTorch).
Experiment Setup Yes For the general setting, the embedding size is 100, the batch size for mini-batch is 100, and the L2 regularization is 10−5. For DHCN, an initial learning rate 0.001 is used. The number of layers is different in different datasets. For Nowplaying and Diginetica, a three-layer setting is the best, while for Tmall, one-layer setting achieves the best performance.