We developed a method of retrieving relevant document elements, exploiting both structural information and the statistics of term distributions in structured hypertext documents. The main thrust of this solution is to allow the retrieval of relevant document elements with arbitrary granularity using keyword-based queries and users can browse the returned documents in their original structures. See the demo.

The system can be browsed in three different ways:

	By searching the corpus using natural language queries.
	By navigating around the hyperdocument using a browser pane on the right that displays the documents.
	By browsing through different parts of documents using the tree view structure displayed on the left pane.