I have using an extra core for Auto complete -- i periodically build
an index where each document is an item I want to appear in the
"suggest" list. I build tokens using the EdgeNGram filter. I have
found that using a special analyzer gives much better results then
simply using facet prefix query.
Somewhere *low* on my TODO list is to look into implementing this with
a prefix ree or whatever my be the best solution. I figure down the
line we could have something that keeps the same API, but does not hit
the index (though with caching etc, maybe it would not anyway?)
so yes. I'm interested - but I think building auto complete off a
field type will give better results then starting with the
FacetComponent.
ryan
On Aug 12, 2008, at 3:13 PM, Shalin Shekhar Mangar wrote:
Hi,
I was reading through some old threads on solr-user on implementing an
auto-complete feature. A lot of times it was suggested that people
should
use in-memory trie data structures to implement this efficiently for
high
query loads rather than using Solr's facet.prefix support, perhaps
at the
cost of additional memory.
How about adding this functionality to Solr itself? I haven't
thought a lot
about on how to implement this but I just wanted to see what you
guys think
about it. It can be a handler or a search component or it can be an
opt-in
enhancement to FacetComponent itself. Just like SpellCheckComponent,
it
would have commands to build the radix tree (or whatever data
structure we
decide on) and to query it. It can give only the suggestions (terms)
or both
suggestions+counts, if the user wants.
Thoughts?
--
Regards,
Shalin Shekhar Mangar.
