Knowledge-Enhanced Hierarchical Graph Transformer Network for Multi-Behavior Recommendation

Authors: Lianghao Xia, Chao Huang, Yong Xu, Peng Dai, Xiyue Zhang, Hongsheng Yang, Jian Pei, Liefeng Bo4486-4493

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments conducted on three real-world datasets show that KHGT consistently outperforms many state-of-the-art recommendation methods across various evaluation settings.
Researcher Affiliation Collaboration 1South China University of Technology, China, 2JD Finance America Corporation, USA 3Communication and Computer Network Laboratory of Guangdong, China 4Peng Cheng Laboratory, Shenzhen, China, 5Simon Fraser University, Canada
Pseudocode No The paper describes the model architecture and methodology in text and figures, but does not include any explicit pseudocode or algorithm blocks.
Open Source Code Yes Our implementation code is available in https://github.com/akaxlh/KHGT.
Open Datasets Yes Data Description. We show the data statistics in Table 1. Movie Lens Data. We differentiate explicit user s rating scores r (i.e., [1, ..., 5]) into multiple behavior types... Yelp Data. User s feedbacck interactions over items in this data, are projected into three behavior types... Online Retail Data. We also investigate our KHGT in a real-world online retail scenario with explicit multi-typed user-item interactions...
Dataset Splits No Following the same settings in (Yu et al. 2019; Zhao et al. 2020), we employ the time-aware leave-one-out evaluation to split the training/test sets. The test set contains the last interactive item of users and the rest of data is used for training.
Hardware Specification No The paper does not provide any specific hardware details such as CPU or GPU models, or cloud computing instance types, used for running the experiments.
Software Dependencies No The paper states 'We implement KHGT with Tensor Flow' but does not provide specific version numbers for TensorFlow or any other software dependencies.
Experiment Setup Yes Our multi-head self-attention module is configured with the 2 heads for embedding learning. The channels of base transformations in our graph attention module is set as 2. The model training process is performed with the learning rate of 1e 3 (with decay rate of 0.96) and batch size of 32. The regularization strategy with the weight decay, which is chosen from the set of {0.1, 0.05, 0.01, 0.005, 0.001}.