Building XML statistics for the hidden web

Abstract

There have been several techniques proposed for building statistics for static XML data. However, very little work has been done in the area of building XML statistics for data sources that export XML views of data that is stored in relational or other databases. For such data sources, we need statistics that are built in an on-line manner, by observing the XML queries to the data sources and their results. In this paper, we present a technique for building on-line XML statistics by observing the XPath queries issued to a data source and their result sizes. These XPath queries select parts of the virtual XML document representing the XML view of the data at the data source. We convert these XPath queries to a more abstract and generalized form that we call annotated path expressions. We present a technique for storing these annotated path expressions and information about their selectivity for use in estimating the selectivity of future XPath queries. We also present an experimental evaluation of our proposed approach. Copyright 2003 ACM.