Synthesis Process Paragraph Extraction from Scientific Literature of Inorganic Material

Abstract

In the field of inorganic materials, there is a need for a system that supports development by analyzing synthesis processes described in a large number of papers. In order to realize the system, it is necessary to extract the part where synthesis processes are described from the papers. We propose a tool that extracts paragraphs describing synthesis processes from papers in the PDF format. We develop the tool by combining a deep learning-based sentence classifier that determines whether each sentence includes synthesis processes or not and a paragraph detector using the sentence classifier. In the experiment, we evaluated our tool on manually-labeled 300 papers. As a result, our tool performed well in both classifying sentences and detecting paragraphs. This result shows that the proposed tool is useful in extracting paragraphs on synthesis processes.