Data quality in web archiving

Abstract

Web archives preserve the history of Web sites and have high long-term value for media and business analysts. Such archives are maintained by periodically re-crawling entire Web sites of interest. From an archivist's point of view, the ideal case to ensure highest possible data quality of the archive would be to "freeze" the complete contents of an entire Web site during the time span of crawling and capturing the site. Of course, this is practically infeasible. To comply with the politeness specification of a Web site, the crawler needs to pause between subsequent http requests in order to avoid unduly high load on the site's http server. As a consequence, capturing a large Web site may span hours or even days, which increases the risk that contents collected so far are incoherent with the parts that are still to be crawled. This paper introduces a model for identifying coherent sections of an archive and, thus, measuring the data quality in Web archiving. Additionally, we present a crawling strategy that aims to ensure archive coherence by minimizing the diffusion of Web site captures. Preliminary experiments demonstrate the usefulness of the model and the effectiveness of the strategy.