Semistructured graph-based databases have been proposed as well-suited
stores for World-Wide Web data. Yet so far, languages for querying
such data are too complex for casual Web users. Further, proposed
query approaches do not take advantage of the interactivity of typical
Web sessions--users are proficient at iteratively refining their Web
explorations. In this paper we propose a new model for interactively
querying and searching semistructured databases.  Users can begin with
a simple keyword search, dynamically browse the structure of the
result, and then submit further refining queries. Enabling this model
exposes new requirements of a semistructured database that are not
apparent under traditional database uses. We demonstrate the
importance of efficient keyword search, structural summaries of query
results, and support for inverse pointers. We also describe some
preliminary solutions to these technical issues.