RESEARCHarXiv CS.CL·5d ago
MM-BizRAG: Rethinking Multimodal Retrieval-Augmented Generation for General Purpose Enterprise Q&A
MM-BizRAG proposes a direct approach for multimodal retrieval-augmented generation in enterprise Q&A, explicitly handling structured information in complex documents. It uses a document structure-aware split and orientation-specific ingestion pipelines to better process various document types.
29