AttributeLevel Neighbor Hierarchy Construction Using Evolved PatternBased Knowledge
Attribute-Level Neighbor Hierarchy Construction Using Evolved Pattern-Based Knowledge Induction Author : Thanit Puthpongsiriporn, J. David Porter, Bopaya Bidanda, Ming-En Wang, and Richard E. Billo Reporter : Tze Ho-Lin 2006/10/4 TKDE, 2006 1
Outline n n n n Motivation Objectives Method PKI & e. PKI Experimental result Concluding remarks Appendix Personal Comments 2
Motivation n n Original Pattern-based Knowledge Induction (PKI) constructs neighbor hierarchies at the tuple level. If a query is issued to retrieve a nonexisting tuple such as an aluminum stock with WIDTH=w 1, LENGTH=l 1, and THICKNESS=t 5, the neighbor hierarchy depicted in Fig. 1 will fail to generate approximate answers. Fig. 1 3
Objectives n To propose the Evolved PKI (e. PKI) construct attribute value neighbor hierarchies at both the tuple and attribute levels utilizing inferential relationships among attributes in the relation 4
Method: PKI Preliminaries P PA PB A Set & A Value Normalization 5
Method: e. PKI n: the number of unique attributes in the relation p is 0 if A is also a unique attribute; otherwise, p equals 1 and m-n-p!=0 6
Experimental result Grouping Efficiency Chandrasekharan and Rajagopalan’s (ZODIAC) best block diagonal form: 0. 6933 e. PKI-based block diagonal transformation algorithm: 0. 732 7
Concluding remarks n The e. PKI technique allows for the construction of neighbor hierarchies for nonunique attributes based upon confidences, popularities, and correlations of relationships among attribute values. 8
Appendix Grp Eff 9
Personal Comments n Application q n Advantage q q n Document, Website, Patent, etc. Fault Tolerance: the e. PKI could retrieve a nonexisting tuple to generate approximate answers. The e. PKI is applicable for both categorical and numerical attribute values. Disadvantage q The higher the number of irrelevant, the lower the resulting nearness values. 10
Neighbor Hierarchy Table 2 Relation ALUM_PLATE 11
- Slides: 11