The volume of academic literature, such as academic conference papers and journals, has increased rapidly worldwide, and research on metadata extraction is ongoing. However, high-performing metadata extraction is still challenging due to diverse layout formats according to journal publishers. To accommodate the diversity of the layouts of academic journals, we propose a novel LAyout-aware Metadata Extraction (LAME) framework equipped with the three characteristics (e.g., design of automatic layout analysis, construction of a large meta-data training set, and implementation of metadata extractor). In the framework, we designed an automatic layout analysis using PDFMiner. Based on the layout analysis, a large volume of metadata-separated training data, including the title, abstract, author name, author affiliated organization, and keywords, were automatically extracted. Moreover, we constructed a pre-trained model, Layout-MetaBERT, to extract the metadata from academic journals with varying layout formats. The experimental results with our metadata extractor exhibited robust performance (Macro-F1, 93.27%) in metadata extraction for unseen journals with different layout formats.

New scientific and technological (S&T) knowledge is being introduced rapidly, and hence, analysis efforts to understand and analyze new published S&T documents are increasing daily. Automated text mining and vision recognition techniques alleviate the burden somewhat, but the various document layout formats and knowledge content granularities across the S&T field make it challenging. Therefore, this paper proposes LA-SEE (LAME and Vi-SEE), a knowledge graph construction framework that simultaneously extracts meta-information and useful image objects from S&T documents in various layout formats. We adopt Layout-aware Metadata Extraction (LAME), which can accurately extract metadata from various layout formats, and implement a transformer-based instance segmentation (i.e., Vision based Semantic Elements Extraction (Vi-SEE)) to maximize the vision-based semantic element recognition. Moreover, to constructing a scientific knowledge graph consisting of multiple S&T documents, we newly defined an extensible Semantic Elements Knowledge Graph (SEKG) structure. For now, we succeeded in extracting about 6 million semantic elements from 49,649 PDFs. In addition, to illustrate the potential power of our SEKG, we provide two promising application scenarios, such as a scientific knowledge guide across multiple S&T documents and questions and answering over scientific tables.