Gathering metadata from Web-based repositories of historical publications

Building digital libraries from Internet-accessible document repositories is a challenging task, due to the current mismatch between the desired DBMS-like capabilities of the former and the schemeless HTML files stored in web sires. In order to address this problem, we propose a distributed architecture for the extraction of metadata from WWW documents specially suited for repositories of historical publications, like newspapers. In this paper we present an information extraction system based on semi-structured data analysis. Starting from several combinations of the HTML styles that abstract the visual characteristics of documents, the proposed system infers the logical structure and attributes of HTML texts. Additionally, by using context-free grammars the system extracts the overall web structure of the repositories. The system output is a metadata object that contains a concise representation of the corresponding publication and its components