Dynamic Taxonomies (FIND Workshop)

1 minute read

The term dynamic taxonomies refers to a multidimensional (multifaceted *check*) classification .

Interesting ideas:

Searching/Browsing: express selections as queries; provide a A-box of the current selection, which acts as the base for further manipulations
DT-miner: computes co-occurrences using taxonomies yielding a higher (ontology dependent) abstraction level.
```
    beer  -> beverages
    wine  -> beverages
    the   -> article
    chips -> snacks
  
```
Possible ontologies: WordNet et. al
concepts are not terms; but the author uses a pragmatic approach and defines them as label for multidimensional classes.
Term expansion: shallow vs. deep (see graphic); advantage: only consider the current part of the search space (shallow/deep)
related documents: establish a relationship between two documents if both of them share at least one common classification. => empirical evidence Application: e.g. the wine taxonomy

Searches: show a box with the domains of the first 20 search results; allow usage of these domains as filter criteria by (de)selection of single domains
Query extension: apply a reasoner to the T-Box to extend queries; serialize the reasoner's results to improve the process's performance.

by Banek, Marko

Idea: every class is expressed as WordNet synset tree => compare both trees wordnet sense $$w_i$$ "most similar" to $$w_j$$ => meaning

Literature:
- semantic similarity calculation for WordNet (Yank & Powers, 2005)
- classic disambiguation techniques (Liu, Yu & Meng 2005)