Research map / Large language models / AI and society
ESG Accountability Made Easy: DocQA at Your Service
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Presents Deep Search DocQA, a system that converts ESG PDF reports to machine-readable form and answers questions over them with a retrieval-plus-LLM conversational assistant.
Problem
Information in ESG disclosure reports is mostly locked in non-machine-readable PDFs, making it hard for researchers, policy-makers, and companies to extract data and insights.
Approach
An end-to-end pipeline combining computer-vision document conversion (layout analysis and table structure extraction), embedding-based retrieval over text and table triplets stored in a vector database, and LLM response generation grounded in retrieved context with safety and hallucination checks.
Key results
- Deep Search DocQA indexes over 10,000 ESG disclosure reports from over 2,000 corporations for question answering.
- Table 1 shows example question answers over IBM 2022 and Starbucks 2022 ESG reports, but no quantitative evaluation is reported.
Limitations
- No quantitative evaluation of answer accuracy or retrieval quality is reported.
- Currently answers questions over single documents only; multi-document querying is future work.
Experimental setup
Datasets: ESG disclosure reports (over 10,000 reports from over 2,000 corporations)
Builds on: Retrieval Augmented Generation Sentence Transformers
Closest papers by meaning
- Benchmarking Retrieval-Augmented Multimomal Generation for Document Question Answering (NeurIPS 2025 · similarity 0.81)
- DocMIA: Document-Level Membership Inference Attacks against DocVQA Models (ICLR 2025 · similarity 0.81)
- M3Grounder: Mask-Based Multi-Span and Multi-Granular Grounding for Document QA (CVPR 2026 · similarity 0.81)
- CVQA: Culturally-diverse Multilingual Visual Question Answering Benchmark (NeurIPS 2024 · similarity 0.80)
- GRAM: Global Reasoning for Multi-Page VQA (CVPR 2024 · similarity 0.80)
- RepLiQA: A Question-Answering Dataset for Benchmarking LLMs on Unseen Reference Content (NeurIPS 2024 · similarity 0.80)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.