A rule-based information extraction system for human-readable semi-structured scientific documents

The explosion of data over the past twenty years has fostered a huge amount of research in processing semi-structured documents like HTML and XML documents on Web. Nevertheless, the explosion of semi-structured documents that originate from outside the Web domain is more challenging. The data of semi-structured documents are everywhere: in scientific research reports, official journals, electronic health records and any records from other domains. Different from HTML and XML, semi-structured record is usually human-readable and has its own internal schema. On account of the incredible increase of data volumes, the traditional methods of studying on these data cannot meet the needs of high performance and flexible access. Relational database technique is a good alternative technique for managing and organizing data. In this paper, we aim at (1) revealing the semantic structure in human-readable scientific records (HSR), and (2) presenting a transformation system that converts HSR into structured relational database. Our approach identifies the relevant logical sub-structures and objects in HSR according to specific templates. We don´t target “the best” methods but provide an available framework for facilitating the semi-structured HSR analysis tasks. The evaluation to the system shows its ability to enable the efficient transformation of HSR and therefore contributes to the increasing corpus of semi-structured HSR documents.