Project 2 Ontology alignment

1
Project 2Ontology alignment
2
(No Transcript)
3
Ontology Alignment

• Defining the relations between the terms in
  different ontologies

4
An Alignment Framework
5
Matcher Strategies

• Strategies based on linguistic matching
• Structure-based strategies
• Constraint-based approaches
• Instance-based strategies
• Use of auxiliary information

6
Aim

• Gain understanding about the ontology alignment
  process
• Gain understanding about about advantages and
  disadvantages of different strategies for
  ontology alignment matching strategies and
  single threshold filtering
• Gain understanding about evaluation of strategies
  using precision, recall, f-measure
• Learn about the Ontology Alignment Evaluation
  Initiative (OAEI)
• Learn to use the tools and data sets of the OAEI

7
Tasks

1. Tutorial - learn how to use the tools provided by
   the OAEI (matching, single threshold filtering,
   evaluation)
2. Run existing algorithms on the benchmark test and
   discuss results
3. Implement own matcher, evaluate and discuss
   results
4. (study other test cases and discuss what kind of
   matchers would be appropiate)

8
Task 2 using task 1 data

• Method Precision Recall F-score
  Threshold
• SMOA 0.69 0.96
  0.80 0.5
• SMOA 0.73 0.96
  0.83 0.6
• SMOA 0.79 0.96
  0.87 0.7
• SMOA 0.80 0.94
  0.87 0.8
• SMOA 0.88 0.73
  0.80 0.9
• Levenshtein 0.80 0.94
  0.87 0.5
• Levenshtein 0.86 0.77
  0.81 0.6
• Levenshtein 0.92 0.46
  0.61 0.7
• Levenshtein 0.94 0.35
  0.52 0.8
• Levenshtein 0.99 0.35
  0.52 0.9

9
Task 3

• Implementation of own matchers.
• Definition of similarity computation
• Testing using thresholds

10
Task 3
11
Task 3
12
Task 3
13
Task 3 performance on task1 data

• Method Precision Recall F-score
  Threshold
• equal 1.00 0.23
  0.37 1
• SMOA 0.69 0.96
  0.80 0.5
• Levenshtein 0.68 0.98
  0.80 0.33
• Levenshtein 0.53 1.00
  0.69 0
• BOWOverlap 0.87 0.85 0.86
  0.3
• WordOverlap 1.00 0.60 0.75
  0.3
• BOTOverlap 0.76 0.85 0.80
  0.3
• TrigramOverlap 0.80 0.92 0.85
  0.1
• FinalScore 0.87 0.94
  0.90 0.2
• NEW 0.48 0.92
  0.63 0
• NEW 0.74 0.88
  0.80 0.88

14
Task 2 - benchmark

• Benchmark
• 1xx (4) same ontology, no overlap, language
  generalization, language restriction
• 2xx (ca 40) base ontology with modified base
  ontology (e.g. change names, remove relations,
  spelling mistakes, use of sysnonyms, change of
  natural language)
• 3xx (4) real cases

15
Task 2 benchmark data

• Best system OAEI 2007 p 0.95 r 0.9
• Range p 0.98 (with r 0.64) - 0.76 (with r
  0.7)
• r 0.90 (with p 0.95) - 0.21 (with
  p 0.92)
• According to category
• 1xx several systems p1, r1
• 2xx p 0.95, r 0.9 p 0.97, r 0.89
• 3xx p 0.94, r 0.68

16
Task 2 - benchmark

• Approximate string matching
• (WordNet)
• Multilingual WordNet
• Structure
• Instances

17
Task 2/3