rLLM: Relational Table Learning with LLMs

TL;DR

The paper presents rLLM, a modular PyTorch framework for Relational Table Learning (RTL) that decomposes Graph Neural Networks, Large Language Models, and Table Neural Networks into standardized modules, enabling rapid construction of RTL-type models via a combine-align-co-train workflow. It introduces BRIDGE, a simple RTL method combining a Table Encoder and a Graph Encoder to model multi-table relational data with reduced architectural complexity, along with three enhanced relational datasets (TML1M, TLF2K, TACM12K) under the SJTUTables collection. Experimental results on TML1M show BRIDGE achieving superior accuracy over strong single-table baselines, illustrating the benefit of cross-table information fusion. The work provides a practical, extensible platform for RTL research and encourages standardized evaluation with easy-to-use datasets, potentially accelerating RTL method development and application.

Abstract

We introduce rLLM (relationLLM), a PyTorch library designed for Relational Table Learning (RTL) with Large Language Models (LLMs). The core idea is to decompose state-of-the-art Graph Neural Networks, LLMs, and Table Neural Networks into standardized modules, to enable the fast construction of novel RTL-type models in a simple "combine, align, and co-train" manner. To illustrate the usage of rLLM, we introduce a simple RTL method named \textbf{BRIDGE}. Additionally, we present three novel relational tabular datasets (TML1M, TLF2K, and TACM12K) by enhancing classic datasets. We hope rLLM can serve as a useful and easy-to-use development framework for RTL-related tasks. Our code is available at: https://github.com/rllm-project/rllm.

Figures (3)

• Figure 1: The architecture of rLLM
• Figure 2: Base data structure in rLLM. Arrows indicate inheritance relationships and parentheses indicate containment relationships.
• Figure 3: The architecture of BRIDGE