Albert Weichselbraun

is a Professor of Information Science at the University of Applied of the Grisons.

Remarks on Ontology Learning and Evaluation

less than 1 minute read

This post contains some random remarks on ontology learning and evaluation:

  • terms versus concepts: concepts are formed by grouping terms with the same meaning
  • collocation = co-occurrence of a sequence of words (significant phrase detection)
  • kinds of similarity: (i) paradigmatic similarity - terms are substitutable for each other in a given context (e.g. carrot and onion as different kinds of vegetables) (ii) syntagmatic similarity - there is an association between the terms (relatedTo; e.g. "car" and "drive") which can be detected by statistical means (e.g. co-occurrence)

  • limited paradigmatic modifiability for detecting collocations (Wermter & Hahn 2005)
    • easy to substitute words in a collocation candidate with other words (= high paradigmatic modifiability) => unlikely that the collocation is correct
    • limited modifiability => high probability that the collocation is correct

Approaches for Evaluating Ontology Learning

Source: Dellschaft & Staab 2008

  • task-based evaluation
  • corpus-based evaluation - compare ontology concepts with corpus concepts
  • criteria-based evaluation - based on criteria suggested by the domain experts (e.g. how many terms where aggregated to form a concept, interconnectivity, ...)

You may also enjoy

Suffix array

1 minute read

The suffix array is a memory-efficient alternative to the suffix tree which provides a sorted list of string indices indicating the string’s suffixes.