Building a Recommendation Engine: The XELOPES Library - Realtime Data Mining - page 232

12.1 The XELOPES Library
12.1.1 The Main Design Principles
12.1.1.1 Basis Transformations
XELOPES uses a radical concept to solve analysis tasks: basis transformations!
The main task of XELOPES is to provide appropriate bases and efficient basis
transformations in order to solve data analysis problems efficiently.
The term basis is used in a wider sense than of a mathematical basis only. Basis
transformations are applied on the following three levels:
1. Basis transformations of data : For analysis tasks, data is often required to be
presented in different formats, i.e., in different bases. There exist different bases
of training and application data, transactional and non-transactional formats,
dense and sparse formats, etc.
2. Basis transformations inside analysis algorithms : Most analysis algorithms can
be described and implemented in terms of basis transformation methods; we
discussed this in Sect. 6.1 .
3. Basis transformations into CWM basis: The CWM standard (see below) which is
used as fundament of XELOPES is a basis itself! Thus, the data that usually
exists in many proprietary data storage formats can be represented in the unified
CWM format.
Basis transformations consist of two parts:
1. Find an appropriate basis for a problem.
2. Find an efficient basis transformation into the new basis.
Both problems are usually not easy to solve. However, the most important step is
to understand that such a basis transformation is actually necessary. This requires
separating the basis from the data (more specifically, the coordinates of the data
instances). Most systems in data mining simply forgo such separation. In contrast,
XELOPES clearly distinguishes between basis and the data itself. This separation is
done on all three levels mentioned. It is absolutely fundamental for understanding
XELOPES.
12.1.1.2 Modular Concept: CWM
CWM (Common Warehouse Metamodel) is the fundament of XELOPES. The
primary objective of the CWM is to define a metamodel of a generic data ware-
house architecture. In combination with MOF-related standards like XMI and JMI,
CWM allows to exchange metadata between BI applications of different vendors
and types. In particular, metadata between data warehouses can be exchanged.