A Provable Framework of Learning Graph Embeddings via Summarization
Authors: Houquan Zhou, Shenghua Liu, Danai Koutra, Huawei Shen, Xueqi Cheng
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Through extensive experiments on real-world datasets, we demonstrate that our methods can learn graph embeddings with matching or better performance on downstream tasks. |
| Researcher Affiliation | Academia | 1 Institute of Computing Technology, Chinese Academy of Sciences 2 University of Chinese Academy of Sciences 3 University of Michigan |
| Pseudocode | Yes | Algorithm 1: GELSUMM-Deep Walk/LINE/GCN |
| Open Source Code | No | Explanation: The paper references source code for existing methods (LINE, HARP, MILE, Graph Zoom) and a summarization algorithm (Zhou et al. 2021) which is a component, but does not explicitly state that the source code for the proposed GELSUMM framework itself is openly available or provide a link to it. |
| Open Datasets | Yes | Datasets used in our experiments contain three citation networks, two social networks, and one co-purchasing network. The statistics of them are listed in Table 4. ... We use the widely-used dataset splits as in (Kipf and Welling 2017; Hamilton, Ying, and Leskovec 2017). |
| Dataset Splits | Yes | We use the widely-used dataset splits as in (Kipf and Welling 2017; Hamilton, Ying, and Leskovec 2017). |
| Hardware Specification | Yes | All experiments are conducted on a machine with a 2.40 GHz Intel Xeon E5-2640 CPU, a Tesla K80 GPU, and 128 GB RAM. |
| Software Dependencies | No | We implement Deep Walk in Python and use LINE s code released by the authors2. Explanation: While Python is mentioned as the implementation language, no specific version of Python or other libraries/packages with their versions are listed. |
| Experiment Setup | Yes | Hyperparameters: Following (Perozzi, Al-Rfou, and Skiena 2014), in Deep Walk, we sample 10 |V| random walk sequences. The walk length is set to 80 and the window size is set to 10. In LINE, we set the number of negative samples to 5. The embedding dimension is set to 128 for all methods. For GELSUMM-Deep Walk and GELSUMM-LINE, k is set to 2. ... For GELSUMM-GCN, the learning rate is set to 0.001 and the dropout rate is set to 0.5. k is set to 1 on all datasets except Flickr and Pubmed (k is set to 2). Training epoch is set to 200 for all datasets. |