In this paper, we propose the graph neural network (GNN)-based matrix completion technique to reconstruct the de-sired low-rank matrix by exploiting the underlying graph structure of the matrix. The proposed approach, referred to as GNN-based low-rank matrix completion (GNN-LRMC), combines the GNN and the neural-network weight update mechanism. The GNN is used to extract the node vectors of the graph using a modified convolution operation. Empirical simulations validate the reconstruction performance of GNN-LRMC in synthetic and Netflix datasets.
Paper
Full text
Low-Rank Matrix Completion Using Graph Neural Network
Semantic Scholar · Computer Science · 2020
Abstract
In this paper, we propose the graph neural network (GNN)-based matrix completion technique to reconstruct the de-sired low-rank matrix by exploiting the underlying graph structure of the matrix. The proposed approach, referred to as GNN-based low-rank matrix completion (GNN-LRMC), combines the GNN and the neural-network weight update mechanism. The GNN is used to extract the node vectors of the graph using a modified convolution operation. Empirical simulations validate the reconstruction performance of GNN-LRMC in synthetic and Netflix datasets.