J. Diederich, W.-T. Balke,Automatically Created Concept Graphs Using Descriptive Keywords in the Medical Domain,Methods Inf Med 3/2008,pp.241-250
The authors pointed out that existing categorization systems suffer from large manual effort and cost and the Semantic Growbag algorithm fully automatically creates concept graphs, i.e. directed graphs similar to categorization systems but without strong subsumption semantics. The basic idea of the Semantic GrowBag algorithm is to create concept graphs from a corpus of objects annotated with descriptive keywords including hidden relationships between individual keywords, which is done by exploiting higher-order co-occurrence.
The Semantic GrowBag algorithm comprises three main parts:
(1) Compute a new co-occurrence metric based on higher-order co-occurrence
(2) Find relationships between keywords, based on this metric
(3) Construct for each keyword i a GrowBag graph to present a limited view on the‘neighborhood’ of i (i.e., closely related (non-hierarchically subsumed) keywords+ subsumeda keywords).
Moreover, the authors compared Semantic GrowBag concept relationships to the MeSH thesaurus, and found that the vast majority of GrowBag relationships showing a hierarchical character are never contradicting MeSH relationships.
Semantic GrowBag algorithm allows to create hierarchical relationships between meaningful keywords to construct concept graphs without requiring additional information. And Semantic GrowBag algorithm can be used for the maintenance of ontology and terminology systems including MeSH to deal with the ever-evolving medical knowledge.
The authors plan to conduct a user study to actually show the benefits of Semantic GrowBag algorithm while using them, such as document retrieval. Also, the authors plan to extend their approach into the popular collaborative tagging domain like flickr.