Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-Scale Heterogeneous Text-Attributed Graph Datasets From Diverse Domains

Yunhui Liu, Qizhuo Xie, Jinwei Shi, Jiaxu Shen, Tieke He

TL;DR

The paper tackles the lack of diverse, real-world benchmarks for heterogeneous text-attributed graphs (HTAGs) by introducing six multi-domain HTAG datasets (TMDB, CroVal, ArXiv, Book, DBLP, Patent) with original texts, time-based splits, and an automated evaluation pipeline. It demonstrates that heterogeneous graph neural networks generally outperform homogeneous baselines across these datasets, highlighting the value of modeling both text and heterogeneous structure. The work provides detailed dataset construction, baselines, and reproducible code, enabling broad benchmarking and future exploration of joint PLM-HGNN methods and additional tasks beyond node classification. Overall, the HTAG benchmark suite offers a scalable, domain-spanning resource to advance HTAG learning and model generalization in real-world settings.

Abstract

Heterogeneous Text-Attributed Graphs (HTAGs), where different types of entities are not only associated with texts but also connected by diverse relationships, have gained widespread popularity and application across various domains. However, current research on text-attributed graph learning predominantly focuses on homogeneous graphs, which feature a single node and edge type, thus leaving a gap in understanding how methods perform on HTAGs. One crucial reason is the lack of comprehensive HTAG datasets that offer original textual content and span multiple domains of varying sizes. To this end, we introduce a collection of challenging and diverse benchmark datasets for realistic and reproducible evaluation of machine learning models on HTAGs. Our HTAG datasets are multi-scale, span years in duration, and cover a wide range of domains, including movie, community question answering, academic, literature, and patent networks. We further conduct benchmark experiments on these datasets with various graph neural networks. All source data, dataset construction codes, processed HTAGs, data loaders, benchmark codes, and evaluation setup are publicly available at GitHub and Hugging Face.

Multi-Scale Heterogeneous Text-Attributed Graph Datasets From Diverse Domains

TL;DR

The paper tackles the lack of diverse, real-world benchmarks for heterogeneous text-attributed graphs (HTAGs) by introducing six multi-domain HTAG datasets (TMDB, CroVal, ArXiv, Book, DBLP, Patent) with original texts, time-based splits, and an automated evaluation pipeline. It demonstrates that heterogeneous graph neural networks generally outperform homogeneous baselines across these datasets, highlighting the value of modeling both text and heterogeneous structure. The work provides detailed dataset construction, baselines, and reproducible code, enabling broad benchmarking and future exploration of joint PLM-HGNN methods and additional tasks beyond node classification. Overall, the HTAG benchmark suite offers a scalable, domain-spanning resource to advance HTAG learning and model generalization in real-world settings.

Abstract

Heterogeneous Text-Attributed Graphs (HTAGs), where different types of entities are not only associated with texts but also connected by diverse relationships, have gained widespread popularity and application across various domains. However, current research on text-attributed graph learning predominantly focuses on homogeneous graphs, which feature a single node and edge type, thus leaving a gap in understanding how methods perform on HTAGs. One crucial reason is the lack of comprehensive HTAG datasets that offer original textual content and span multiple domains of varying sizes. To this end, we introduce a collection of challenging and diverse benchmark datasets for realistic and reproducible evaluation of machine learning models on HTAGs. Our HTAG datasets are multi-scale, span years in duration, and cover a wide range of domains, including movie, community question answering, academic, literature, and patent networks. We further conduct benchmark experiments on these datasets with various graph neural networks. All source data, dataset construction codes, processed HTAGs, data loaders, benchmark codes, and evaluation setup are publicly available at GitHub and Hugging Face.

Paper Structure

This paper contains 27 sections, 1 equation, 1 figure, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Datasets for Text-Attributed Graphs
  4. Learning on Text-Attributed Graphs
  5. Heterogeneous Graph Neural Networks
  6. Preliminaries
  7. Heterogeneous Text-Attributed Graphs
  8. Heterogeneous Graph Neural Networks
  9. Datasets
  10. TMDB: Movie Collaboration Network
  11. CroVal: Community Question Answering Network
  12. ArXiv: Academic Network
  13. Book: Book Publication Network
  14. DBLP: Academic Network
  15. Patent: Patent Application Network
  16. ...and 12 more sections

Figures (1)

  • Figure 1: As an instance of the heterogeneous text-attributed graph, an academic network contains three types of nodes (papers, authors, and fields of study) with textual descriptions as well as their multi-relational connections (a paper “cites” another paper, an author “writes” a paper, and a paper “has a topic of” a field of study).