From: Li Yan (lanti@u.washington.edu)
Date: Sun May 02 2004 - 22:32:58 PDT
Information Integration Using Logic Views
It is always interesting to read J.D.Ullman's writings, be
it academic paper or textbook, and this one is no exception.
Information Integration is certainly an interesting problem
with great practical value, and it's even more interesting
to discover the underling priciple of the two proposed
approaches are both based on containment of conjuctive query
(CQ) and datalog programs.
Both IM and Tsimmis concentrate on CQ as the model of both
data and queries, and this decision is reasonable given CQ
simple to work with and there are theorems giving bounds on
searching minimum solution. Both are based on logical views
of data sources. However, they differ in the interface
exported. IM uses a set of global predicates, and both the
views and queries are defined in terms of these
predicates. A search on the views finds the minimal
solutions. The solution is the union of all minimal
solutions. Tsimmis answer queries through mediators. The
mediator approach is inflexible when compared with the view
approach in IM, because when a query is posed while all the
available mediators are inconvient, then the resulting
solution is awkard, but in general, queries are arbitrary
and we have to make a compromise in how many mediators we
are to support. Also a large number of mediators are hard to
maintain when new data source were introduced, but IM's
views can remain unchanged except adding a new set of
views. This paper didn't mention support of semistructured
data in IM, but I guess there should be no problem, since IM
didn't have a notion about type even ...
There are other difference between the two systems, but the
striking part is that both of them were based on logical
views, an ancient theory finds its way in a modern
application.
This archive was generated by hypermail 2.1.6 : Sun May 02 2004 - 22:33:00 PDT