Logo image
Back
Docs2Synth: A Synthetic Data Tuned Retriever Framework for Documents Understanding
Conference proceeding   Open access

Docs2Synth: A Synthetic Data Tuned Retriever Framework for Documents Understanding

Yihao Ding, Qiang Sun, Puzhen Wu, Sirui Li, Siwen Luo and Wei Liu
Companion Proceedings of the ACM Web Conference 2026, pp.124-127
ACM Conferences
WWW '26: The ACM Web Conference 2026 (Dubai, United Arab Emirates, 29/06/2026–03/07/2026)
2026
DOI: https://doi.org/10.1145/3774905.3793117
pdf
Published2.79 MBDownloadView
Open Access CC BY-NC-ND V4.0

Abstract

Information systems -- Information retrieval
Document understanding (VRDU) in regulated domains is particularly challenging, since scanned documents often contain sensitive, evolving, and domain specific knowledge. This leads to two major challenges: the lack of manual annotations for model adaptation and the difficulty for pretrained models to stay up-to-date with domain-specific facts. While Multimodal Large Language Models (MLLMs) show strong zero-shot abilities, they still suffer from hallucination and limited domain grounding. In contrast, discriminative Vision-Language Pre-trained Models (VLPMs) provide reliable grounding but require costly annotations to cover new domains. We introduce Docs2Synth, a synthetic-supervision framework that enables retrieval-guided inference for private and low-resource domains. Docs2Synth automatically processes raw document collections, generates and verifies diverse QA pairs via an agent-based system, and trains a lightweight visual retriever to extract domain-relevant evidence. During inference, the retriever collaborates with an MLLM through an iterative retrieval--generation loop, reducing hallucination and improving response consistency. We further deliver Docs2Synth as an easy-to-use Python package, enabling plug-and-play deployment across diverse real-world scenarios. Experiments on multiple VRDU benchmarks show that Docs2Synth substantially enhances grounding and domain generalization without requiring human annotations.

Details

Metrics

2 Record Views
Logo image