Need some tips on suggest / autocomplete


(Benoît) #1

Hi,

I've search the ML about suggest / auto-complete.

The best way to implement this seems to use a multi-field with an edgeNgram
analyzer.
Ok. But how do you build the suggested words list ?

Any tips ?

Benoît

--


(Itamar Syn-Hershko) #2

Lucene 4 is just around the corner, I'd wait for it as it has better
support for this

On Mon, Oct 8, 2012 at 6:04 PM, Benoît benoit.intrw@gmail.com wrote:

Hi,

I've search the ML about suggest / auto-complete.

The best way to implement this seems to use a multi-field with an
edgeNgram analyzer.
Ok. But how do you build the suggested words list ?

Any tips ?

Benoît

--

--


(Chris Male) #3

The EdgeNGram approach analyzes content as it goes into the index, forming
the grams which are used to find suggestions. Consequently the 'suggested
word list' is whatever you put into the index. This could either be your
actual indexed content, or it could be an index you created specifically to
provide suggestions.

On Tuesday, October 9, 2012 5:04:40 AM UTC+13, Benoît wrote:

Hi,

I've search the ML about suggest / auto-complete.

The best way to implement this seems to use a multi-field with an
edgeNgram analyzer.
Ok. But how do you build the suggested words list ?

Any tips ?

Benoît

--


(Otis Gospodnetić) #4

Hi Benoît,

Have you looked at https://github.com/spinscale/elasticsearch-suggest-plugin (based
on Lucene's FST suggester)?

People who use our suggester (
http://sematext.com/products/autocomplete/index.html) tend to feed field(s)
from their main index as suggestions, e.g., product names, book author
names, etc. If you look at http://search-lucene.com/ you'll see we do
something similar there - Wiki page titles, JIRA issues, email subjects,
etc. are all fed into AC and used for suggestions.

We've helped clients feed their queries into AC, too, but this involves a
lot of query cleanup, normalization (think N queries that are all unique,
but really only slightly different), ranking, etc.

Otis

Search Analytics - http://sematext.com/search-analytics/index.html
Performance Monitoring - http://sematext.com/spm/index.html

On Monday, October 8, 2012 12:04:40 PM UTC-4, Benoît wrote:

Hi,

I've search the ML about suggest / auto-complete.

The best way to implement this seems to use a multi-field with an
edgeNgram analyzer.
Ok. But how do you build the suggested words list ?

Any tips ?

Benoît

--


(Benoît) #5

Hello all,

Thank you for your replies.

I've looked at the "Suggester Plugin for Elasticsearch" but this is not for
me : big warning "THIS IS NOT PRODUCTION READY! DO NOT USE IT." , plus the
fact that it need to be rebuild periodically and work at the level of one
index.

With the edgeNgram field I’ve configured and some tuning of highlighting
configuration I’m able to retrieve a list of word easily cleanable but the
problem is that there is many times same word in results.

I will look to implement suggest on my side or with a dedicated index.

Benoît

--


(system) #6