Enhancement: fuzziness above 2 edits

[jguay]

Currently elasticsearch support fuzziness of up to two edits per documentation

Requested feature is for fuzziness of 3 (or more) edits

Example with ALEJANDRO is indexed but returned when ALEXANDER or ALEJATYPO is searched with fuzziness of 3 Edits.

Side notes :

• Synonyms which would be a valid/cheaper solution in some cases which is outside the scope of the feature request
• With async search elasticsearch currently supports better running long queries. Fuzziness of 3 will obviously be much more costly than 2 edits (potentially doc update needed if the feature for 3 or more max edits is implemented)

[elasticmachine]

Pinging @elastic/es-search (Team:Search)

[markharwood]

I wasn't close to the development of the fuzzy implementation but it was notoriously complex getting beyond just one edit distance. That suggests it's not as simple as raising a setting.
It's worth mentioning that a common and perhaps simpler solution to fuzzy matching is to index with ngrams and search those.
Or did you mean that for edit distance > 2 we should only use async search and revert to a slower brute force scan of indexed terms (using Levenshtein string comparisons) rather than adding to the complex automaton-based matching exclusively used today?

[jguay]

thank you @markharwood for confirming this is a difficult feature to implement, the ngram solution is in fact a common good solution although in this case it isn't suitable because mostly it will cause a lot of false positive and the requirement in this case is to always use 3 Edits for the search
Example ALEJANDRO is indexed with "min_gram" : 3 will not match ALZJAZDRZ with 3 Edits which should match and will match ALEZZZZZZ which should not match as this is 6 Edits

I strike that side note on async search, with the expected exponential costs of adding the 3rd Edits, I just wanted to highlight async search might help because the search would run slower if we ever implement max Edit of 3 (when compared to current 2 Edits)

[javanna]

This has been open for quite a while, and hasn't had a lot of interest. For now I'm going to close this as something we aren't planning on implementing. We can re-open it later if needed.