Metadata Extraction from Office Documents

Abstract

This paper focuses on using layout-based techniques to automatically extract metadata when scanning office documents to an archive. Many office documents such as letters, inter-office memos, and invoices contain key information that is spatially arranged. Information arrayed in this manner is easy for a reader to identify and understand. However, location of information within office documents varies greatly between documents, unlike forms where layout is static. This poses a challenge for layout based metadata extraction techniques. Our system uses regular expression matching and stochastic grammars on lines of text to efficiently and accurately label text according to function, enabling archived documents to be precisely retrieved.