Feature Based Approaches to Semantic Similarity

1
Feature Based Approaches to Semantic Similarity

• Kate Deutsch
• May 1, 2008

2
THE BASICS
3
Why feature based??
4
Metric Distance vs. Feature Matching

• Metric distance
• Minimality
• Symmetry --gt --gt
• Triangle Inequality --gt --gt
  then
• 
  --gt
• Feature Matching
• Matching
• Monotonicity
• Independence

5
Assumptions Examined

• Matching
• Similarity
• f(intersection and individual features)
• Monotonicity
• Similarity increases with the addition of common
  features and/or deletion of distinct features
• Independence

6
Matching Functions

• Contrast Model Similarity measurement is a
  linear combination of the measures of common and
  distinctive parts
• Ratio Model Similarity measurement is
  constructed from various set theories and
  normalized

7
Asymmetry and Focus

• Are these the same???
• Assess the degree to which a and b are similar to
  each other
• Assess the degree to which a is similar to b
• Case studies
• Countries
• Figures
• Letters
• Signals

8
What do we do?

• Nevertheless, the symmetry assumption should
  not be rejected altogether. It seems to hold in
  many contexts, and it serves as a useful
  approximation in many others. It cannot be
  accepted, however as a universal principle of
  psychological similarity.
• Can we think of an instance??

9
Feature Similarity and Context
The altering of clusters changes the similarity
of objects in each cluster- diagnosticity
hypothesis
10
Diagnostic Value

• Features that are shared by all objects under
  consideration cannot be used to classify these
  objects and are therefore devoid of diagnostic
  value
• What do you think??

11
MEASURING SIMILARITY
12
LULC systems
Modified Anderson Classification System
National Vegetation Classification System
Elk Habitat Classification System
Attributes, Functions and Parts
Formation of Universe of Discourse
13
(No Transcript)
14
LULC lessons

• Ability for matching is dependent on the need.
• Specificity of matches varies by circumstances (
  Elk shelter vs. Elk food).

15
Geospatial Entities

• Matching-Distance Similarity Measure

Assess Similarity
Distance based
Feature based
Distinguishing Features (attributes, functions,
parts)
Semantic Structure (is-a, part-whole)
16
Geospatial Entities

• Matching process
• Weights defined for the similarity values of
  parts, functions and attributes
• For each type of distinguishing feature,

17
(No Transcript)
18
Applying Weights
19
Similarity Calculation
20
(No Transcript)