TabGraphs: A Benchmark and Strong Baselines for Learning on Graphs with Tabular Node Features

2024-09-22

arXiv.org

AI Summary

Existing graph learning benchmarks are severely limited by narrow domain coverage and overreliance on a few citation networks, hindering the transfer of graph foundation models to diverse real-world applications. Method: We introduce the first benchmark for heterogeneous tabular-feature graphs—comprising 12 real-world business scenarios with mixed numerical/categorical node attributes and structural relationships—and conduct systematic evaluation across GNNs, XGBoost/LightGBM, MLPs, and various feature encoding and graph augmentation strategies. Contribution/Results: We find that GNNs yield only marginal average AUC gains (+1.8%); in contrast, lightweight graph-aware feature engineering—e.g., k-NN neighborhood aggregation coupled with target encoding—enables XGBoost to outperform GNNs on six tasks, with a maximum improvement of +2.3% AUC. This work bridges the gap between tabular and graph learning, highlights the pivotal role of feature engineering in graph-structured tabular tasks, and establishes a new paradigm for fair, practical graph model evaluation and deployment.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community
• Reasoning under Uncertainty: Graphical Models

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Tabular machine learning is an important field for industry and science. In this field, table rows are usually treated as independent data samples, but additional information about relations between them is sometimes available and can be used to improve predictive performance. Such information can be naturally modeled with a graph, thus tabular machine learning may benefit from graph machine learning methods. However, graph machine learning models are typically evaluated on datasets with homogeneous node features, which have little in common with heterogeneous mixtures of numerical and categorical features present in tabular datasets. Thus, there is a critical difference between the data used in tabular and graph machine learning studies, which does not allow one to understand how successfully graph models can be transferred to tabular data. To bridge this gap, we propose a new benchmark of diverse graphs with heterogeneous tabular node features and realistic prediction tasks. We use this benchmark to evaluate a vast set of models, including simple methods previously overlooked in the literature. Our experiments show that graph neural networks (GNNs) can indeed often bring gains in predictive performance for tabular data, but standard tabular models also can be adapted to work with graph data by using simple feature preprocessing, which sometimes enables them to compete with and even outperform GNNs. Based on our empirical study, we provide insights for researchers and practitioners in both tabular and graph machine learning fields.

Problem

• Evaluating graph ML models on narrow academic datasets lacking industrial diversity
• Assessing graph foundation models' transferability across varied real-world domains
• Investigating temporal distribution shifts and GBDT baselines in industrial settings

Innovation

• GraphLand benchmark with 14 diverse industrial datasets
• Evaluates GNNs against gradient-boosted decision trees with graph features
• Tests graph foundation models under realistic temporal distribution shifts