A Novel Construction of Heuristic Coordinator in KDD* Process Model
1
Zhen Peng,
2
Lifeng Wu,
3
Bingru Yang
1
Department of Computer, North China Institute of Science and Technology
2
Information Engineering College, Capital Normal University
3
School of Information Engineering, University of Science and Technology Beijing
wooleef@gmail.com
Abstract
KDD* process has stronger theoretical foundations and better performance than KDD process.
Heuristic coordinator as a key unit plays most important role in KDD* process model. In order to improve
the heuristic coordinator algorithm, a kind of RBFCM model and it’s five important inference mechanisms
are applied in the novel heuristic coordinator. The experiment in protein secondary structure prediction
demonstrates it’s efficiency.
Keywords: KDD*, Heuristic Coordinator, RBFCM(Rule Based Fuzzy Cognitive Map), Protein
Secondary Structure Prediction
1. Introduction
Knowledge Discovery in Database(KDD)
[1,2]
gives an effective way to solve the difficult that data in all
works of life are rich but knowledge is poor. However the research in KDD has mostly been concentrated
on good algorithms for various tasks. Relatively little research has been published about the theoretical
framework or foundations of KDD. To overcome the limitation of weak theoretical foundations and
improve the performance of KDD greatly, we has proposed a new KDD process model-KDD*
[3,4]
, which
regards knowledge discovery as a cognitive system, incorporates double bases cooperating mechanism
[5,6]
constructing 1-1 mapping between databases and knowledge bases and two coordinators (heuristic
coordinator and maintenance coordinator) into classical KDD process model, and improves some mining
methods.
Heuristic coordinator
[3,4]
, a most key unit in KDD*, simulates the “creating intent” of a cognitive
process, is able to automatically find knowledge shortage, inspires the corresponding substructure in the
database, starts the data mining and makes the self focus in order to directional mine in database, and
determines the efficiency and the intelligence of KDD systems. But the proposed one directed hyper-graph
based heuristic coordinator has some flaws such as only taking into account the rules with a single back
part, only containing the existing co-nodes in the knowledge, limited reasoning mechanism and so on.
RBFCM
[7,8]
is a soft computing methodology and has stronger knowledge representation and inference
ability than FCM
[9,10]
. Thus, it is suitable to uses RBFCM to represent knowledge and obtain accessible
matrix by inference mechanism for discovering knowledge shortage and directional mining in the
corresponding database. The paper uses one RBFCM model and the inference mechanisms to improve
heuristic coordinator algorithm in KDD* process model
At last, in order to validate the RBFCM based heuristic coordinator algorithm, we apply it in protein
secondary structure prediction and get better prediction effect.
The rest of the paper is organized as follows: the section 2 introduces the KDD* process model
and heuristic coordinator; knowledge representation based on RBFCM, the section 3 represents
RBFCM based knowledge representation, inference rules. Section 4 evaluates the method in
protein secondary structure prediction. Finally, we conclude this study in Section 5.
2. Heuristic coordinator in KDD* process model
In order to improve the performance of data mining greatly, we originally brought forward to a new
research of inner cognitive mechanism based data mining, whose core idea is to regard the process of data
mining as a process of cognizing, the knowledge discovery system as cognitive system, and form a new
process model- KDD* shown in Figure 1.
A Novel Construction of Heuristic Coordinator in KDD Process Model
Zhen Peng, Lifeng Wu, Bingru Yang
International Journal of Advancements in Computing Technology(IJACT)
Volume4, Number3, February 2012
doi: 10.4156/ijact.vol4.issue3.22