Elasticsearch: Fuziness ignored when multi_match query type is set to cross_fields.

[afoucret]

Is there any reason for this behavior ?
It can be very usefull to use fuziness on cross_fields multi_match queries.

[clintongormley]

Hi @afoucret

Yes, fuzziness really complicates the logic with the blended term query that cross_fields uses internally, and has a significant performance impact. Currently we have no plans to add support for fuzziness to cross_fields.

[afoucret]

Really sad to heard this.

Did you have an alternative or some magic tricks to obtains some this kind of feature ?

[clintongormley]

You can use copy_to to make a custom _all field and then the query becomes a simple match query with fuzziness

[afoucret]

Hi @clintongormley,

One of the shortcoming of the method you propose is that it seems impossible to have weighted fields with copy_to. The other one is the need to reindex all when you want to change the fields used into the query.

I propose the issue stays open because the cross_fields fuzziness is an important feature IMHO.

From what I experienced during my tests, the problem seems to be into blended() which does not support the Lucene fuzziness operator :

The query string I have used fro my test :

blended("bath", field: [name_fr, description_fr])             <-- works
blended("bath~0.2", field: [name_fr, description_fr])       <-- does not works

Am I right ?
If so, is this a ES patch or a Lucene problem (I can find where blended is implemented yet)

[Kaidanov]

I have a query with should with function_score with fuzziness 0.75 with multi_match with cross_field
My query is a term without one last character - for instance "NAM" instead of "NAME"
By fuzziness rules it must be returned to me , but I don't get results.
What would you suggest on this matter?

[ebuildy]

What about throw a QueryParsingException error? I lost 2 hours before arriving on this issue ;-(

Thanks

[markwalsh-liverpool]

+1, would be really handy

[linkwoman]

+1

[cpitzak]

+1 I need this :)

[Kaidanov]

Best_fields worked for me
On Jun 16, 2015 9:30 PM, "Clinton J. Pitzak" [email protected]
wrote:

+1 I need this :)

—
Reply to this email directly or view it on GitHub
https://github.com/elastic/elasticsearch/issues/6866#issuecomment-112522104
.

[tegansnyder]

:+1:

[ctassparker]

:+1:

[PatrickHuetter]

+1

[joshyaugustine]

Would love to see fuzziness support for cross_field queries for the same reason afoucret mentioned earlier in this thread.

[AliHichem]

+1

[MAMABA]

+1

[phber]

+1

[nemopohle]

+1

[vladimirnech]

+1

[kirill-latish]

+1

[deksss]

+1

[exleys]

+1

[DaerBR]

+1

[olehmalkov]

+1

[andrewmaximoff]

+1

[antonydandrea]

+1

[andreaCava]

+1

[kemar]

+1

[jbronen]

+1

[fuzzy-coder]

+1

[AbdullaevAPo]

+1

[kkevilus]

+1

[micseydel]

+1

[anna-kalina]

+1

[zhangjustin]

+1

[flo-ryan]

+1

[exentrich]

+1

[paisanos]

+1

[abhishek0058]

+1

[subwai]

+1 for the same reasons afoucret explained.

My workaround right now is to add another match query and wrap the two in a bool.should clause. This new match query would only match the field i wish to boost with an and operator. This means that if the input is only terms from the boosted field, it will score above all fields which matched via the _all field.
However, if I add more terms so that the boosted field no longer matches all those terms, all fields will be weighted equally.
This solution is good enough for me.