System learns how to best create graphs from text and tables
TACO-Graph: Automated Construction of Text-Atributed Graphs
Machine Learning
Summary
Graphs that represent text data are important for many machine learning tasks, but it’s tricky to know how to build these graphs correctly. The authors created TACO-Graph, a system that tries out different ways to build graphs from text and table data, then quickly scores and chooses the best options using simple tests. This helps people avoid trial and error by providing a clear way to pick the best graph structure for a specific task. It works well on different tasks like classifying documents or retrieving information from product reviews or research papers.
What this means in practice
- •For machine learning engineers: Use TACO-Graph to automatically find the best way to turn tabular text data into graphs for better model performance on classification or retrieval tasks.
- •For data platform developers: Integrate TACO-Graph to optimize graph construction from multi-attribute datasets, improving downstream analytics in citation networks or product review systems.
Authors
Amiri Hayes, Kristina Wicke
Abstract
Graph-structured representations of text data have become foundational across machine learning, yet it remains unclear how to systematically construct a graph from a given dataset for a target task. Choices about which entities become nodes, which relationships become edges, and what text each node contains can substantially affect downstream performance. In practice, finding optimal constructions often requires expert knowledge or manual trial and error. We introduce TACO-Graph, a decision tree framework for learning which graph construction choices are most effective for a given dataset and task. Specifically, we focus on text-attributed graphs (TAGs) constructed from multi-attribute tabular data. TACO-Graph enumerates valid TAG representations, scores each using fast proxy evaluation on small representative subsets, and trains an interpretable decision tree on those scores. Across five datasets spanning citation networks, product reviews, and bibliographic relationships, our decision trees predict construction quality on separately drawn evaluation samples with mean Spearman $ρ$ above $0.9$ for node classification and above $0.8$ for GraphRAG retrieval. Our trees systematically identify which constructions perform better across both GNN and GraphRAG tasks, and the full pipeline is released as an open-source library to support evaluation on new tasks and datasets at https://github.com/AmiriHayes/text_attributed_graphs. TACO-Graph provides a systematic and interpretable guide for determining how to convert tabular data to text-attributed graphs for downstream tasks.