harder to retrieve the integrated subtrees (like the
documents and interrelated tuples) connected by the
content nodes and complementary nodes that do not
contain any keyword but contain some relevant and
meaningful data, since it is rather difficult to determine
which nodes are complementary nodes. For example,
in Fig. 1, if a user inputs the keywords “XML, 2006,
VLDB”, he expects to get the subtree circled by the
double dotted lines as the answer, with the two authors
related to the answer as complementary nodes relevant
to the keyword query. Therefore, this paper discusses
how to retrieve those compact, integral subtrees, called
self-integral trees (SI-Trees), which contain comple-
mentary nodes that capture the focus of keyword que-
ries, besides the content nodes. More importantly, each
self-integral tree represents an integrated meaning to