In this paper, we propose an entity-based style of queries for
semistructured data. First, we partition a semistructured data into
subgraphs corresponding to real-world entities, in other words, into
composite objects. To detect composite objects in semistructured
data, we use the exclusiveness of references. If a reference is
exclusive, then we regard it as a composite link. Then, we develop a
query language for entity-based queries. That language supports path
expressions, in which we can use edge expressions that match only with
composite links or non-composite links. By using these expressions in
combination with wild cards, we can specify queries of a form like
“retrieve all entities including these data items,” which we call
entity-based style queries. We show examples demonstrating how this
style of query is useful especially when one does not have enough
knowledge on the schema in advance.