Motif-Matching Based Subgraph-Level Attentional Convolutional Network for Graph Classification
Authors: Hao Peng, Jianxin Li, Qiran Gong, Yuanxin Ning, Senzhang Wang, Lifang He5387-5394
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results on both bioinformatics and social network datasets show that the proposed models significantly improve graph classification performance over both traditional graph kernel methods and recent deep learning approaches. |
| Researcher Affiliation | Academia | 1Beijing Advanced Innovation Center for Big Data and Brain Computing, Beihang University, Beijing 100191, China 2State Key Laboratory of Software Development Environment, Beihang University, Beijing 100191, China 3Key Laboratory of Aerospace Network Security, Ministry of industry and information technology, School of Cyberspace Science and Technology, Beihang Universty, Beijing 100191, China 4Department of Computer Science, Brown University, Providence, RI, USA 5College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, Nanjing 211106, China. 6Department of Computer Science and Engineering, Lehigh University, Bethlehem, PA, USA |
| Pseudocode | No | The paper describes the proposed methods using text and figures, but it does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | The code of this work is publicly available at https://github.com/RingBDStack/MA-GCNNs. |
| Open Datasets | Yes | We use the following five bioinformatics datasets MUTAG, PTC, PROTEINS, D&D and NCI1. MUTAG is a dataset of 188 mutagenic aromatic and heteroaromatic nitro compounds (Debnath et al. 1991)... PTC (Toivonen, Srinivasan, and Helma 2003) is a dataset of 344 organic molecules... PROTEINS is a graph collection obtained from (Borgwardt, Cheng, and Vishwanathan 2005)... D&D is a dataset of 1178 protein structures (Dobson and Doig 2003)... NCI1 dataset is chemical compounds screened for activity against non-small cell lung cancer and ovarian cancer cell lines (Wale, Watson, and Karypis 2008)... we also select five social network datasets (Yanardag and Vishwanathan 2015), including IMDB-BINARY (IMDB-B), IMDB-MULTI (IMDB-M), REDDIT-BINARY (RE-B), REDDIT-MULTI-5K (REM-5K) and REDDIT-MULTI-12K (RE-M-12K). |
| Dataset Splits | Yes | All of our experiments are evaluated using the 10-fold cross-validation method. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., CPU/GPU models, memory) used for running the experiments. It only mentions 'multi-threaded parallel processing'. |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers (e.g., programming language versions, library versions, or specific framework versions) used for the implementation. |
| Experiment Setup | Yes | The common parameters of training the models are set as MOMENTUM = 0.9, Dropout = 0.5, learning rate = 0.001, and L2 norm regularization weight decay = 0.01. We set F1 = 128, F2 = 64 in M-GCNN model and F = 16, S = 8 in MA-GCNN model. For each dataset, the parameters N, K, w1, w2, w3 and training epochs are set based on the following principles: (1) N is set to the average number of nodes for all the graphs in a given graph dataset. (2) The numbers of nodes K in the subgraph are set to 10 and 20 in bioinformatics and social network datasets, respectively. (3) The numbers of w1, w2, w3 are set based on the subgraph connectivity information. Considering the number of training sample and downward trend of the objective function, we adjust the batch size from 45 to 450 to get the best accuracy. |