Stabilizing and Enhancing Link Prediction through Deepened Graph Auto-Encoders

Authors: Xinxing wu, Qiang Cheng

IJCAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirically, extensive experiments on various datasets demonstrate the competitive performance of our proposed approach. Theoretically, we prove that our deep extensions can inclusively express multiple polynomial filters with different orders.
Researcher Affiliation Academia University of Kentucky, Lexington, Kentucky, U.S.A. xinxingwu@gmail.com, qiang.cheng@uky.edu
Pseudocode No The paper describes the models using mathematical equations but does not include any pseudocode or algorithm blocks.
Open Source Code Yes The codes of this paper are available at https://github.com/xinxingwu-uk/DGAE.
Open Datasets Yes Firstly, we employ three standard benchmark datasets, i.e., Cora, Citeseer, and Pubmed; then, we also evaluate our deep extensions on three webpage-related datasets. We summarize the data statistics in Table 1.
Dataset Splits Yes We train all the models by randomly removing 15% of links while keeping all node features, and the validation and test sets are formed by a ratio of 5:10 from the removed edges and the corresponding node pairs; the models weights are initialized by using the Glorot uniform technique. For all experiments, the obtained mean results and standard deviations are for 10 runs over 10 different random train/validation/ test splits of datasets.
Hardware Specification No The paper does not specify the hardware used for running the experiments (e.g., CPU, GPU models, memory, or cloud instances).
Software Dependencies No The paper mentions using the Adam optimizer and t-SNE but does not provide specific version numbers for any software libraries or frameworks used in the implementation.
Experiment Setup Yes In all experiments, we set the maximum number of epochs to 200 and adopt the Adam optimizer with an initial learning rate of 0.01. For simplicity, we construct our deep encoders with (k 1) 32-neuron hidden layers for k in (5) or (6) and a 16-neuron latent embedding layer. Besides, we perform a grid search in {0.000001, 0.000005, 0.00001, 0.00005, 0.0001, 0.0005, 0.001, 0.005, 0.01, 0.05, 0.1, 0.5, 1, 1.5, 2} to tune hyper-parameters for our models according to the performance on the validation set.