Ignore the word order in the search query

[weiziqian]

At the moment, the order of query words matters a lot. For example:

1. Searching for "XXX Log Cabin 50mm" returns nothing if none of the document contains the first query word "XXX", even though many documents contain "Log Cabin 50mm".
2. Searching for "Log Cabin XXX 50mm" will return documents containing "Log Cabin", but the ranking completely ignores whether "50mm" is present. For instance, "A Log Cabin 40mm" receives the same score as "B Log Cabin 50mm".

(ps. in the 2nd example, because 'matchingStrategy' is default to 'last', the "50mm" will be the first word to remove from the query, and then the "XXX". such that "50mm" makes no effect to the ranking score)

Would be great if as an option, the words in the search query can be treated equally.
Thanks!

[weiziqian]

An alternative to this, we can use "matchingStrategy" in the search api.
It currently only supports: last / frequency / all.
Should be another option : "infrequent", to remove query words from the most infrequent word in the database.

[ManyTheFish]

Hey @weiziqian, did you try the frequency matching strategy?

[weiziqian]

Hi @ManyTheFish , thanks for the reply!

1. Using frequency does not help. For example, with the keyword "XXX Cabin 40mm", Meilisearch removes "Cabin" first and then "40mm". However, "XXX" has no (or very few) matches, so it stays. This means it doesn’t return more results containing "Cabin" or "40mm".
2. I made a small hack in the Meilisearch code to reverse the removal order based on frequency, and it works perfect. (for example, the remove order becomes "XXX" -> "40mm" -> "Cabin", with each removal it gives more results)

[ManyTheFish]

Hello @weiziqian,

Sorry for the delay of my answer,

unfortunatelly, I don't think that, removing infrequent word first, would really help.
The main reason is that a searh engine is made to retrieve a subset of documents that are the most relevant for the query. However, keeping only the frequent words would not help to tie the documents and would retrieve most of the dataset for each query.
For instance, let's say we have an address dataset and we search for "13 backer street", an infrequent matching strategy would end by retrieving all the documents containing "street" which is basically all the dataset.

[weiziqian]

Thanks @ManyTheFish
I see what you mean, however, the problem I've mentioned, is not that the search result is too much, but is too few.
often because in the search term, there is an "non-exist" word. and this word won't be removed because the system thinks it is the most infrequent word.

And that's why (when search result is too few) we need this 'matchingStrategy' to remove search terms. we don't need it in your example.