[Corpora-List] software for measuring semantic similarity and relatedness?

John F Sowa sowa at bestweb.net
Sun Oct 6 17:39:55 UTC 2013


On 10/6/2013 12:52 PM, Alexander Osherenko wrote:
> I considered measuring semantic similarity using findings in PhD of
> Alexander Budanitsky. Here is a brief introduction -- Budanitsky,
> Alexander and Hirst, Graeme, Evaluating WordNet-based Measures of
> Lexical Semantic Relatedness (dl.acm.org/citation.cfm?id=1168108
> <http://dl.acm.org/citation.cfm?id=1168108>).

That's a good summary of the issues.  From page 32:

> a word should be considered non-anomalous in the context of another if
> there is any kind of semantic relationship at all apparent between them.
> ...
> As mentioned in the introduction, these can include just about any kind
> of functional relation or frequent association in the world.

There is a way to find such relationships:

  1. Translate every document in a corpus to a semantic network in
     any of the many notations that have been used over the years.

  2. Store all the graphs in a system that enables any query graph
     to retrieve all similar graphs within a given semantic distance.

To be practical, this method requires a system that allows similar
graphs to be retrieved in time proportional to the logarithm of the
number of graphs in the store. For issues and citations, see slides
129 to 132 of http://www.jfsowa.com/talks/goal.pdf

The remaining slides show some applications that use such a store.

John

_______________________________________________
UNSUBSCRIBE from this page: http://mailman.uib.no/options/corpora
Corpora mailing list
Corpora at uib.no
http://mailman.uib.no/listinfo/corpora



More information about the Corpora mailing list