Interoperability and Integration of Processes of Knowledge Discovery
in Databases (Teza de Doctorat, coordonator: prof.dr. Toader
Jucan)
-----------------------------------
          
ABSTRACT
-----------------------------------
In the context of interoperability in knowledge 
discovery in databases, this thesis proposes an 
architecture of an online scoring system (De-Visa) 
that can be integrated easily in loosely coupled 
service oriented architectures. 
It manages a repository of PMML models as a native
XML database and exploits their predictive or 
descriptive properties.
This thesis proposes a novel technique for online 
scoring based on web services and the specification 
of a specialized XML-based query language for PMML 
models called PMQL, used to enable communication 
with the data mining consumers and for processing 
the models in the repository. 
At the abstract level, the thesis presents an 
theoretical foundation that captures both structural 
behavioral aspects of the system providing solutions 
to problems that arise. The structural aspects includes 
the mining models/schemas, the data dictionaries etc. 
The behavioral aspects include the way the system 
interacts with consumers requests, namely scoring or 
composition requests. A theoretical framework for 
allowing prediction model composition is provided. 
Among others it uses the concept of semantic 
consequence in the functional dependencies theory. 
In the context of online scoring a novel hybrid technique 
for schema matching is provided. The technique is based 
on a modified version of the cycle canceling max-flow 
min-cost algorithm that allows integrating additional 
constraints such as derivability and validity. 
It also proposes an adaptive similarity measure based 
on string metrics, Jaccard index in the textual description, 
field statistics and lexical sense. In this context the 
work presents the global data dictionary architecture 
that alleviates the overhead of matching complexity 
within the scoring process and an algorithm for online 
incremental update of the GDD.
          
SPEAKER(S)
-----------------------------------
Asist.dr. Diana GOREA
Universitatea Alexandru Ioan Cuza, Facultatea de Informatica
Iasi
Romania
-----------------------------------