MMESGBench: Pioneering Multimodal Understanding and Complex Reasoning Benchmark for ESG Tasks

TL;DR

MMESGBench is a pioneering benchmark dataset designed to evaluate multimodal understanding and complex reasoning in ESG documents, addressing the challenges posed by their diverse structure and multimodal content.

Contribution

It introduces the first comprehensive ESG benchmark with multimodal QA pairs, created through a human-AI collaborative pipeline, enabling better evaluation of AI models in this domain.

Findings

Multimodal models outperform text-only baselines on the benchmark.

Models show particular strength on visually grounded and cross-page questions.

The dataset covers diverse ESG document types and source categories.

Abstract

Environmental, Social, and Governance (ESG) reports are essential for evaluating sustainability practices, ensuring regulatory compliance, and promoting financial transparency. However, these documents are often lengthy, structurally diverse, and multimodal, comprising dense text, structured tables, complex figures, and layout-dependent semantics. Existing AI systems often struggle to perform reliable document-level reasoning in such settings, and no dedicated benchmark currently exists in ESG domain. To fill the gap, we introduce \textbf{MMESGBench}, a first-of-its-kind benchmark dataset targeted to evaluate multimodal understanding and complex reasoning across structurally diverse and multi-source ESG documents. This dataset is constructed via a human-AI collaborative, multi-stage pipeline. First, a multimodal LLM generates candidate question-answer (QA) pairs by jointly interpreting rich textual, tabular, and visual information from layout-aware document pages. Second, an LLM verifies the semantic accuracy, completeness, and reasoning complexity of each QA pair. This automated process is followed by an expert-in-the-loop validation, where domain specialists validate and calibrate QA pairs to ensure quality, relevance, and diversity. MMESGBench comprises 933 validated QA pairs derived from 45 ESG documents, spanning across seven distinct document types and three major ESG source categories. Questions are categorized as single-page, cross-page, or unanswerable, with each accompanied by fine-grained multimodal evidence. Initial experiments validate that multimodal and retrieval-augmented models substantially outperform text-only baselines, particularly on visually grounded and cross-page tasks. MMESGBench is publicly available as an open-source dataset at https://github.com/Zhanglei1103/MMESGBench.