KMail Intelligently reading emails...

[Scott Wheeler]

On Sunday 20 March 2005 2:15, John Tapsell wrote:
> Hey klink guys,
>   I'd be interested in any thoughts in the same-named email thread on
> kde-devel ( KMail Intelligently reading emails...)  in regard to klink..
> Particulary about whether a realtime full text parser for kmail (and
> others) would be useful for feeding straight into klink.

Well, first Till (one of the main KMail folks) and I have talked a lot about 
the use of KLink in KMail since, well, we're sharing an office at the moment 
at work.

I don't think the sort of hard coded keyword approach that you're going after 
there will be the best way to approach things.  More of the KLink-style 
approach would be something like:

*) Enter both words into a search of some sort -- i.e. "John" and "phone" (or 
if it's specifically a dialog for looking for a certain type of information 
-- say a phone number one or some of those words might already be provided by 
the application)

*) Using those words as starting nodes in the graph begin traversing nodes via 
links (graph edges) to see if there's a path that links them in the graph

*) Use a proximity metric gathered from the indexer to rank the results -- 
this is still a bit ill defined, but the basics are that proximity is a mix 
of distance in the graph and proximity in a given resource

*) Report most relevant results, if any

Static semantics like hard coded keywords are something generally to be 
avoided in these sorts of structures though.  For instance your patch is very 
English specific.

-Scott

-- 
If you want to get laid, go to college, but if you want an education, go to 
the library.
--Frank Zappa