Database Reference
In-Depth Information
Figure 3. Inference by automatic constraint propagation
In the general case, operations of projection and de-projection can be combined so that it has a zig-
zag form and consists of upward (projection) and downward (de-projection) segments in the partially
ordered structure of collections (Savinov, 2005a, 2005b).
Inference
Logical navigation using projection and de-projection can be used for constraint propagation along
explicitly specified dimension paths. In this section we describe an inference procedure which can
au-
tomatically
propagate source constraints through the model to the target without complete specification
of the propagation path (Savinov, 2006b). This inference procedure is able to choose a natural propaga-
tion path itself. This means that if something is said about one set of elements then the system can infer
something specific about other elements in the model. The main problem is
how
concretely the source
constraints are propagated through the model.
In COM, the inference procedure consists of two steps:
1. De-projecting source constraints down to the chosen fact collection
2. Projecting the obtained fact elements up to the target collection
Let us consider a concept-oriented database schema shown in Figure 3 where it is assumed that each
topic is published by one publisher and there is a many-to-many relationship between writers and topics.
If we want to find all writers of one publisher then this can be done by specifying concrete access path
for propagating publishers to writers:
(Publishers | name = “XYZ”)
<-
published
<-
(Books)
<-
book
<-
(BooksWriters)
->
writer
->
(Writers)
Here we select a publisher, de-project it down to the Books collection, then again de-project the
result further down to the BooksWriters collections with fact elements, and finally project the facts up
to the Writers collection.
