Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Article Classification with Graph Neural Networks and Multigraphs

Khang Ly, Yury Kashnitsky, Savvas Chamezopoulos, Valeria Krzhizhanovskaya

TL;DR

The paper tackles the challenge of classifying scholarly articles into context-specific taxonomies by enriching Graph Neural Network inputs with multi-graph representations that encode multiple signals of relatedness. It combines References, Authorship, Source, and Subject Area edge types with state-of-the-art LM-based textual embeddings (SimTG, TAPE) and applies an R-GCN transformation to enable heterogeneous inputs, showing consistent performance gains across OGBN-arXiv and PubMed diabetes. The main contributions are a data-driven multi-graph construction methodology, an extensive ablation study identifying robust edge-type configurations, and evidence that simple 2-layer GNNs can achieve competitive results with SOTA methods when fed enriched graphs. This approach offers a scalable, architecture-lean path to improved article classification using readily available metadata and lightweight models, with reproducibility resources provided.

Abstract

Classifying research output into context-specific label taxonomies is a challenging and relevant downstream task, given the volume of existing and newly published articles. We propose a method to enhance the performance of article classification by enriching simple Graph Neural Network (GNN) pipelines with multi-graph representations that simultaneously encode multiple signals of article relatedness, e.g. references, co-authorship, shared publication source, shared subject headings, as distinct edge types. Fully supervised transductive node classification experiments are conducted on the Open Graph Benchmark OGBN-arXiv dataset and the PubMed diabetes dataset, augmented with additional metadata from Microsoft Academic Graph and PubMed Central, respectively. The results demonstrate that multi-graphs consistently improve the performance of a variety of GNN models compared to the default graphs. When deployed with SOTA textual node embedding methods, the transformed multi-graphs enable simple and shallow 2-layer GNN pipelines to achieve results on par with more complex architectures.

Article Classification with Graph Neural Networks and Multigraphs

TL;DR

The paper tackles the challenge of classifying scholarly articles into context-specific taxonomies by enriching Graph Neural Network inputs with multi-graph representations that encode multiple signals of relatedness. It combines References, Authorship, Source, and Subject Area edge types with state-of-the-art LM-based textual embeddings (SimTG, TAPE) and applies an R-GCN transformation to enable heterogeneous inputs, showing consistent performance gains across OGBN-arXiv and PubMed diabetes. The main contributions are a data-driven multi-graph construction methodology, an extensive ablation study identifying robust edge-type configurations, and evidence that simple 2-layer GNNs can achieve competitive results with SOTA methods when fed enriched graphs. This approach offers a scalable, architecture-lean path to improved article classification using readily available metadata and lightweight models, with reproducibility resources provided.

Abstract

Classifying research output into context-specific label taxonomies is a challenging and relevant downstream task, given the volume of existing and newly published articles. We propose a method to enhance the performance of article classification by enriching simple Graph Neural Network (GNN) pipelines with multi-graph representations that simultaneously encode multiple signals of article relatedness, e.g. references, co-authorship, shared publication source, shared subject headings, as distinct edge types. Fully supervised transductive node classification experiments are conducted on the Open Graph Benchmark OGBN-arXiv dataset and the PubMed diabetes dataset, augmented with additional metadata from Microsoft Academic Graph and PubMed Central, respectively. The results demonstrate that multi-graphs consistently improve the performance of a variety of GNN models compared to the default graphs. When deployed with SOTA textual node embedding methods, the transformed multi-graphs enable simple and shallow 2-layer GNN pipelines to achieve results on par with more complex architectures.
Paper Structure (10 sections, 2 figures, 5 tables)

This paper contains 10 sections, 2 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Datasets
  4. Data Augmentation
  5. Subgraph Properties
  6. Experiments and Results
  7. Ablation Study
  8. Optimal Configuration
  9. Conclusions and Future Work
  10. References

Figures (2)

  • Figure 1: Illustration of the proposed multi-graph input, which enables the neighboring feature aggregation for a node $X_1$ to be performed across a variety of subgraphs, leveraging multiple signals of article relatedness (References, Authorship, and shared Journal depicted here).
  • Figure 2: Degree distribution, i.e. frequency of each degree value, of all subgraphs for OGBN-arXiv (left) and PubMed (right), plotted on a log-log scale. Points indicate the unique degree values.