Inductive Matrix Completion Based on Graph Neural Networks

Authors: Muhan Zhang, Yixin Chen

ICLR 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We compare IGMC with state-of-the-art matrix completion algorithms on five benchmark datasets. Without using any content, IGMC achieves the smallest RMSEs on four of them, even beating many transductive baselines augmented by side information.
Researcher Affiliation Collaboration Muhan Zhang* Washington University in St. Louis muhan@wustl.edu *Now at Facebook Yixin Chen Washington University in St. Louis chen@wustl.edu
Pseudocode Yes Algorithm 1 ENCLOSING SUBGRAPH EXTRACTION
Open Source Code Yes Our code is publicly available at https://github.com/muhanzhang/IGMC.
Open Datasets Yes We conduct experiments on five common matrix completion datasets: Flixster (Jamali & Ester, 2010), Douban (Ma et al., 2011), Yahoo Music (Dror et al., 2011), Movie Lens-100K and Movie Lens-1M (Miller et al., 2003).
Dataset Splits Yes For ML-100K, we train and evaluate on the canonical u1.base/u1.test train/test split. For ML-1M, we randomly split it into 90% and 10% train/test sets. For Flixster, Douban and Yahoo Music we use the preprocessed subsets and splits provided by (Monti et al., 2017).
Hardware Specification No The paper does not provide specific details regarding the hardware used for running the experiments, such as GPU or CPU models.
Software Dependencies No The paper states 'We implemented IGMC using pytorch geometric (Fey & Lenssen, 2019)' but does not provide specific version numbers for PyTorch Geometric or any other key software libraries.
Experiment Setup Yes The final architecture uses 4 R-GCN layers with 32, 32, 32, 32 hidden dimensions. Basis decomposition with 4 bases is used... The final MLP has 128 hidden units and a dropout rate of 0.5. We use 1-hop enclosing subgraphs... randomly drop out its adjacency matrix entries with a probability of 0.2... We set the λ in (7) to 0.001. We train our model using the Adam optimizer... with a batch size of 50 and an initial learning rate of 0.001, and multiply the learning rate by 0.1 every 20 epochs for ML-1M, and every 50 epochs for all other datasets.