Leveraging Node Attributes for Incomplete Relational Data

Authors: He Zhao, Lan Du, Wray Buntine

ICML 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments show that our models achieve the stateof-the-art link prediction results, especially with highly incomplete relational data.
Researcher Affiliation Academia 1Faculty of Information Technology, Monash University, Australia. Correspondence to: He Zhao <he.zhao@monash.edu>.
Pseudocode No The paper describes mathematical models and inference steps through equations and text, but it does not contain a structured pseudocode or algorithm block.
Open Source Code Yes 1Code available at https://github.com/ ethanhezhao/NARM/
Open Datasets Yes Lazega-cowork: This dataset (Lazega, 2001) contains 378 links of the co-work relationship among 71 attorneys... NIPS234... (Zhou, 2015)... Facebook-ego: The original dataset (Mc Auley & Leskovec, 2012)... Citeseer: This dataset2 contains a citation network... 2http://linqs.umiacs.umd.edu/projects/ /projects/lbc/index.html... Cora: This dataset2 contains a citation network... Aminer: The Aminer dataset (Tang et al., 2009)... Cora-hier... extracted from the original Cora dataset3. 3https://people.cs.umass.edu/ mccallum/ data.html
Dataset Splits No The paper specifies training data proportions (e.g., 'varied the training data from 10% to 90%') and states 'used the remaining in testing', but it does not explicitly define a separate validation dataset split.
Hardware Specification Yes (all implemented in MATLAB and running on a desktop with 3.40 GHz CPU and 16GB RAM)
Software Dependencies No The paper states 'all implemented in MATLAB' but does not specify a version number for MATLAB or other key software dependencies.
Experiment Setup Yes For Sym-NARM and EPM, we set the truncation level large enough for each dataset: Kmax = 50, 100, 256 for Lazega-cowork, Facebook-ego and NIPS234, NIPS12 respectively... Following (Zhou, 2015), we used 3000 MCMC iterations and computed AUC-ROC/PR with the average probability over the last 1500.