GraphOTTER: Evolving LLM-based Graph Reasoning for Complex Table Question Answering

TL;DR

GraphOTTER introduces a graph-based explicit reasoning approach for complex table question answering, improving accuracy by filtering irrelevant information and constructing clear reasoning paths.

Contribution

It is the first to explicitly model reasoning on tables using graph transformations and step-by-step actions guided by LLMs, enhancing interpretability and effectiveness.

Findings

Outperforms existing methods on benchmark datasets

Effectively filters irrelevant information during reasoning

Demonstrates robustness across different LLM backbones

Abstract

Complex Table Question Answering involves providing accurate answers to specific questions based on intricate tables that exhibit complex layouts and flexible header locations. Despite considerable progress having been made in the LLM era, the reasoning processes of existing methods are often implicit, feeding the entire table into prompts, making it difficult to effectively filter out irrelevant information in the table. To this end, we propose GraphOTTER that explicitly establishes the reasoning process to pinpoint the correct answers. In particular, GraphOTTER leverages a graph-based representation, transforming the complex table into an undirected graph. It then conducts step-by-step reasoning on the graph, with each step guided by a set of pre-defined intermediate reasoning actions. As such, it constructs a clear reasoning path and effectively identifies the answer to a given question. Comprehensive experiments on two benchmark datasets and two LLM backbones demonstrate the effectiveness of GraphOTTER. Further analysis indicates that its success may be attributed to the ability to efficiently filter out irrelevant information, thereby focusing the reasoning process on the most pertinent data. Our code and experimental datasets are available at \url{https://github.com/JDing0521/GraphOTTER}.