Monitoring XML data on the Web

We consider the monitoring of a flow of incoming documents. More precisely, we present here the monitoring used in a very large warehouse built from X ML documents found on the web. The flow of documents consists in XML pages (that are warehoused) and HTML pages (that are not). Our contributions are the following: a subscription language which specifies the monitoring of pages when fetche d, the periodical evaluation of continuous queries and the production of XM L reports. the description of the architecture of the system we implemented that makes it possible to monitor a flow of millions of pages per day with millions o f subscriptions on a single PC, and scales up by using more machines. a new algorithm for processing alerts that can be used in a wider context. We support monitoring at the page level (e.g., discovery, of a new page wit hin a certain semantic domain) as well as at the element level (e.g., inser tion of a new electronic product in a catalog). This work is part of the Xyleme system. Xyleme is developed on a cluster of PCs under Linux with Corba communications. The part of the system describe d in this paper has been implemented. We mention first experiments.