Spectral-Based Graph Neural Networks for Complementary Item Recommendation
Authors: Haitong Luo, Xuying Meng, Suhang Wang, Hanyun Cao, Weiyao Zhang, Yequan Wang, Yujun Zhang
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results on four e-commerce datasets demonstrate the effectiveness of our model, with SCom GNN significantly outperforming existing baseline models. |
| Researcher Affiliation | Collaboration | 1Institute of Computing Technology, Chinese Academy of Sciences, 2University of Chinese Academy of Sciences, 3Pennsylvania State University, 4BAAI, 5Purple Mountain Laboratories, 6Nanjing Institute of Infor Super Bahn |
| Pseudocode | No | The algorithm and detailed time complexity analysis can be found in the supplementary files, where the time complexity of spectral-based GCN filters and the two-stage attention mechanism is O(3|E|) and O(8|d |), respectively. |
| Open Source Code | Yes | Detailed descriptions of baselines and implementations can be found in the supplementary files and our code is available1. 1https://github.com/luohaitong/SCom GNN |
| Open Datasets | Yes | Following (Liu et al. 2020; Hao et al. 2020; Bibas, Shalom, and Jannach 2023), we use publicly available benchmark datasets from Amazon. We consider alsobought as complementary relationships, and our task is to realize the link prediction on the complementary item graphs. We select four datasets: Appliances , Grocery , Toys , and Home , and use the categories and price of each item as features. |
| Dataset Splits | Yes | Similar to previous work (Liu et al. 2020), for each item, we randomly sample one edge for constructing the test data, one for the validation data, and use the remaining edges as the training data. |
| 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 provide specific version numbers for software dependencies. |
| Experiment Setup | Yes | For our implementation, we set the embedding size and network layers of both two GCN filters to 16 and 1, respectively. ... where M is the number of the negative items, and τ is a temperature hyperparameter. |