I think the problem is when it combines suggestions from DirectSolrSpellChecker 
and WorkBreakSolrSpellChecker, it gets two lists of possiblities in edit 
distance order.  And when it combines these lists, all it does is interleave 
the 2 lists: 1 from the first list, then 1 from the 2nd list, then 1 from the 
1st, etc.  

So I think if you ran the query with just Direct, you'd see 1 list for each 
potentially misspelled word, and then if you ran the query with just WordBreak, 
you'd see a difference list for each potentially misspelled word.  And then 
when running with both spellcheckers, you'll see them interleaved 
every-other-one.

It might (or might not) depend on the order you specify the 2 spellcheckers in 
solrconfig.xml.  Maybe (not sure here) the first one is guaranteed to provide 
the first suggestion, so long as it provides at least one.  You might want to 
see if you have WordBreak specified first, and if so, then switch them.  
Because when collations are tested, it just goes through the lists, top to 
bottom and tries the various combinations until either "maxCollationTries" or 
"maxCollations" is exhausted.  And it will give you the "good" collations it 
finds in the order it finds them.

Possibly, an easy workaround is to just increase "maxCollations" by 1 more and 
then use the suggestion with the most hits.  This will be a small performance 
penalty though every time it has to find collations, as testing the 
possibilities is expensive.

James Dyer
Ingram Content Group


-----Original Message-----
From: O. Klein [mailto:kl...@octoweb.nl] 
Sent: Tuesday, February 10, 2015 11:55 AM
To: solr-user@lucene.apache.org
Subject: RE: alternativeTermCount and WordBreakSolrSpellChecker combination not 
working

James,

That is very useful information. I tested it and can confirm that disabling
spellcheck in warmer solves core reload problem. 

Now with my use case I'm not trying to spellcheck and correct a whitespace.
If "holy wood" was queried with a mm of 100% it would have fewer hits then
hollywood and this would then be the best correction.

Is there a way to do this?

 



--
View this message in context: 
http://lucene.472066.n3.nabble.com/alternativeTermCount-and-WordBreakSolrSpellChecker-combination-not-working-tp4185352p4185423.html
Sent from the Solr - User mailing list archive at Nabble.com.


Reply via email to