Ontology Learning Mining Functional Dependencies from Data Hong Yao and Howard J. Hamilton Presented By Stephen Lynn.

Ontology Learning

Mining Functional Dependencies from Data

Hong Yao and Howard J. Hamilton

Presented By Stephen Lynn

Ontology Learning

Rule Mining

Algorithmic process that takes data as input and yields rules such as:

Association Rules ImplicationsFunctional dependencies

Ontology Learning

Overview Goals/Objectives Implication/Functional Dependencies Base Algorithm 4 Pruning Rules Evaluation Analysis

Ontology Learning

Goals and Objectives

Design an efficient rule discovery algorithm for mining functional dependencies from a dataset.

Ontology Learning

Implication Describes relationship between one specific

combination of attribute-value pairs.Binary DataPropositional Logic

{milk, eggs} → {bread}

Ontology Learning

Functional Dependency Describe relationship between all possible

combinations of attribute-value pairs.Disjoint attributesTrue regardless of how many possible attribute valuesantecedent → consequent

postcode → areacode

Ontology Learning

Search Space

Ontology Learning

Armstrong’s Axioms

Ontology Learning

Equivalent Attributes

Ontology Learning

Nontrivial Closure

Ontology Learning

Base Algorithm Generate all possible antecedents then test with

possible consequents (1 level at a time)

Ontology Learning

Pruning Rules

Ontology Learning

FD_Mine

Ontology Learning

Experimental Summary 15 Datasets from UCI Machine Learning Repository

(2005)

Ontology Learning

Results

Ontology Learning

Results

Ontology Learning

Runtime

Ontology Learning

Analysis Strengths

Nicely drawn proofs

WeaknessesMissing good exampleNice to show results with/without pruning

Future WorkFind multivalued dependenciesFind conditional dependenciesData cleaning