Research on Document Layout Detection and Description Method for Retrieval-Augmented Generation of Large Language Model

Improve the accuracy of Retrieval-Augmented Generation (RAG) for Large Language Model (LLM), this paper proposes a document layout detection and description network model (Doc-LDNet) for RAG. Doc-LDNet can accomplish two tasks: one is to conduct document layout detection which can accurately identifying the positions and boundaries of different regions in the document, such as text blocks and tables. The other is to describe the images within the document by converting the image content into textual descriptions. Applying Doc-LDNet as a document parsing tool in RAG can effectively enrich the results of document parsing. The experimental results show that applying the Doc-LDNet in RAG can effectively improve the accuracy of LLM in professional domain knowledge question-answering.