Semantic closeness

Semantic proximity is a concept that describes the similarity of documents or terms according to the nature of their content ( meaning , semantics ). Semantic closeness can be illustrated with the help of “maps”, in which similar documents or terms are brought closer together, but dissimilar ones further apart. This happens e.g. B. - often involuntarily - when creating mind maps .

There are also various visualization tools for the web , with the help of which the semantic proximity of web pages (content) is illustrated ( KartOO , WebBrain ).

The definition of semantic proximity is an important problem area when using ontologies for semantic annotation and semantic search, especially in the semantic web .

Formalization of the semantic proximity

Are terms in a tree structure , such as B. in taxonomies , the semantic proximity of two terms can be defined as the length of the shortest path between the term nodes.

If you have information about the frequency of occurrence ( p ) of the hierarchized terms t , e.g. B. from the analysis of a text corpus , these can be quantified by means of their information content I :

${\ displaystyle I (t) = - \ log p (t)}$

A metric for the concept tree can then be defined as follows:

${\ displaystyle \ Delta I (t_ {1}, t_ {2}) = I (t_ {2}) - I (t_ {1})}$ if t _{1 is a} superclass of t ₂ and

${\ displaystyle \ Delta I (t_ {3}, t_ {2}) = \ Delta I (t_ {3}, t_ {1}) + \ Delta I (t_ {2}, t_ {1})}$ if t _{1 is a} common superclass of t ₂ and t ₃ .

Extensions to this simple rule include the density of the tree nodes and the absolute depth in the tree structure.

See also: information theory

Web links

http://www-lehre.informatik.uni-osnabrueck.de/~uwaechte/viz/ Page with visualizations of semantic maps
http://www.touchgraph.com/TGGoogleBrowser.html TouchGraph's Google browser
http://fnps.coli.uni-saarland.de/pathsearch PathFinder through FrameNet (finds shortest paths in the Berkeley FrameNet hierarchy)