Topic similarity measures (in context of Topic Maps)

Hello,

Topic Maps standard specifies prerequisites for merging topics. These 
prerequisites can be though as a topic similarity measure with a binary 
value -- either topics are ultimately similar (1) or ultimately 
different (0).

I would like to ask if anyone has designed/experienced/knows any other 
similarity measures for topics in context of topic maps. I am especially 
interested in similarity measures (and their algorithms) that return not 
binary but real values. I can easily imagine string similarity measures 
such as Levenshtein distance applies well to topic names and URIs but 
how about associations?

Kind Regards,
Aki Kivela
Wandora Team