In any particular combination of domains, although a common understanding o
f the underlying low-level concepts concerning a domain attribute may exist
, different concept hierarchies may have been built at different data-holdi
ng sites. A distributed database may therefore hold different views of the
same data, or differently classified samples from the same population. Beca
use of this, when statistical functions are applied to generate summary tab
les, the resulting summary-based partitions may be heterogeneous. In these
situations, integration of such summary-based partitions can reveal latent
information at a new, and finer, level of granularity. In this paper, the c
lassification schemes are described using a matrix representation of the in
tersection hypergraph, and efficient numerical algorithms are proposed to d
etermine the optimal granularity of the integrated summary data. (C) 1999 E
lsevier Science B.V. All rights reserved.