The implementation assumes that you have 2 entities. In other words we need to suggest text that starts after the separator. What about starting from the 3rd and 4th word? Could we just create a new subfield and change the script for the predicate_token_filter from token.position != 0 to token.position != 0 & token.position != 1 and so on? IMO, could work but there should be a “better” way. Do not forget to set preserve_position_increments as false for the new field. For search time analyzer we want to use a standard analyzers because we should not drop first token from the search string. For indexing we want an analyzer that drops the first token. The idea here is to have a subfield for suggestions that uses different analyzers for indexing and searching. Nothing fancy here, copy-paste from the documentation verbatim, for the sake of completeness. All examples are worked out and tested with version 8.4.1. s -> from the second word trick is to use a different analyser at the search time Ĭool, it’s time to open the Kibana dev tools and hack.b -> the classic use case which already works.You decide to go to the rabbit hole of the analyzers.Īfter the technical refinement meeting the notes on implementation of use cases look like: In the anemic Elasticsearch documentation there is a hint that using text analyzers (stopwords are mentioned) it is possible to achieve different entry points for suggestions which sounds like the trick that might work for our case. Also, the team discards generation of multiple strings using an ingest processor because nobody in our team likes to work with them. Generating multiple strings upstream is not an option due to the time constraints (because in a week the song will not be in the charts anymore). This means and that all b, s, e, j should suggest that artist. The PM specifies that it is needed that artists name and surname should be the source of suggestions. The problematic artist currently rocking in the charts and attracting a lot of attention is Britney Spears & Elton John. You try to argue that it will take forever to change upstream indexing pipeline because it is owned by another department of your company. On top he adds that he doesn’t care whether that makes sense from search engineers perspective, it must be done 2. Of course, the deadline is yesterday, as always.

So, you are a search engineer that happily uses Elasticsearch Completion Suggester feature: lightning speed prefix suggestions works just like a charm 1.īut one day the product manager comes to you with a requirement: could we also suggest if users start typing a word from the middle of the suggested string?.

Elasticsearch text analyzers can supercharge search suggesters.

Completion suggester elasticsearch