Abstract：Vision and Natural Language for Metadata Extraction from Scientific PDF DocumentsA Multimodal Approach The challenge of automatically extracting metadata from scientific PDF documents varies depending on the diversity of layouts within the PDF collectionIn some disciplines such as German social sciencesthe authors are not required to generate their pap Library AutomationWith the continuous expansion of digital library usagethe number of scientific papers published annually in digital format is increasing dramaticallyrequiring automated processing to facilitate paper queriescitation countspaper recommendationsetcRetrieval and IndexingMetadata provides a way for scholars to more easily retrieve and index papersThis article explores the challenges of automatically extracting metadata from scientific PDF documentsDocuments in specific fieldssuch as German social scienceshave diverse appearances and layoutsso using only natural language processingNLPmethods may not be effectiveThereforeresearchers propose a multi-modal neural network model that combines NLP and computer visionCVtechnologies to extract metadataThis method assumes that joint learning of both approaches can better understand document contentthereby improving extraction accuracy