19

Lets say I have following strings indexed:

"My awesome pizza"
"My awesome beer"
"Heineken is awesome beer"
"I love pizza pepperoni"
"Where is my beer"

For input "My" I want to suggest:

  1. My awesome
  2. My beer

For input "awesome" I want to suggest:

  1. awesome pizza
  2. awesome beer

etc.

So I want to provide next words after user input... Also with some fuzziness (aEwsome) should be also supported.

Which suggesters/analyzers I should use? I tried Term and completion but it is not what I want (completion for example works only when search for start of the phrase - if I pass word that is in the middle of indexed string then it will not suggest it,)

user606521
  • 14,486
  • 30
  • 113
  • 204

2 Answers2

18

Just in case anyone bumps in here, the edge-n-gram approach (as noted by @Siddardha Budige) is now superseded by completion suggesters.

They are purposely built with auto-completion in mind, so they are fast end efficient. A very informative post By the ES team can be found here.

Two caveats:

  1. They are implemented by an in-memory Data-Structure (called a Finite State Transducer) and are therefore not suitable for large datasets that can't fit in memory.

  2. The ranking of the results is not based on similarity (a result either matches or not) so you have to be careful with blending fuzzy and non-fuzzy matches.

EDIT: We shared our own experience with using Completion Suggesters here.

It includes some practical examples on fuzzyness and info for non-English inputs.

cmantas
  • 1,516
  • 14
  • 14
7

Elasticsearch provides a convenient way to get autocomplete up and running quickly with its completion suggester feature.

Check this link, this is what you are looking for

https://qbox.io/blog/multi-field-partial-word-autocomplete-in-elasticsearch-using-ngrams

Completion Suggester also helps you to achieve this functionality

cmantas
  • 1,516
  • 14
  • 14
Siddardha Budige
  • 1,005
  • 1
  • 8
  • 12