Auto Complete and search user supplied text/phrase


(nchandra) #1

Hi,

I have a collection of product data (cameras) that I am indexing.
Users can search a product by providing a phrase and I want to enable
'auto complete' or 'auto suggestions' feature.

Being new to search domain, what query types are recommended for these
2 use cases?

Currently, I am inclined towards textPhrasePrefixQuery for 'auto
suggestions' and text query for user provided search string.

Nishant


(freddj) #2

Hi

I'm also implementing an 'autocomplete' feature and I'm not very experienced
in that area.
textPhrasePrefixQuery is a good starting point, but I dropped it for
several reasons:

  • ranking is a bit strange
  • That looked a bit slow to run (but I did not make a scientific
    evaluation of that).
  • if the prefix is too short, it can fail.
  • If you look for 'fo ba', it will miss 'foo bar'

Internally, it generates a lot of sub queries (one for every token in the
index that has the same prefix) and I suppose it does some kind of a massive
OR or dismax between them (that explains the problem I had).

So instead, I choose to index my field with an NGram token filter and I just
do a regular 'text' query on it. I'm happy with the results for now


(system) #3