Facet search array of objects

[rauldeheer]

We have a document structure like so:

{
 ...
 "tags": [
    ...
    { "id": "uuid", "value": "EUR", "type": "CURRENCY" }
    ...
  ]
  ...
}

On our website we show three filter options. You can select one or multiple currencies, platforms (Steam, Amazon, etc) and target countries (Belgium, The Netherlands, etc). So we basically need to show each option in our selection list (uniquely).

Right now we just did a search and gathered the data in code, but once we have a lot of documents this will be an issue, especially considering the fact we can't retrieve 10-20-30 thousand hits at once for every client visiting our website. So we went through the Meilisearch docs and we found facets.

Facets seem to be perfect. We don't need hits so we just set the hitsPerPage on 0 and for the facetName we select tags. Now this works, but it just doesn't return anything. If we then change this value to tags.value, we get all the values of all tags uniquely. This is great. We'll just have to do 3 separate facet searches for each filter, not perfect but better than what we have now.

The problem however is that once we set facetQuery to something like tags.type = CURRENCY, we still get all the other tag types like PLATFORM and COUNTRY.

How can we achieve what we want? What would be the best solution here and does Meilisearch even support this?

[Kerollmops]

Hey @rauldeheer 👋

I understand your issue a little bit, but only some things. It seems to be related to this one in some way where a user wants to filter documents by using multiple fields of an object inside the main document.

Facets seem to be perfect. We don't need hits so we just set the hitsPerPage on 0 and for the facetName we select tags. Now this works, but it just doesn't return anything.

That's what made me think about the above issue. Meilisearch flattens the subobjects in a document. It's like the objects in tags see their fields explode and an array of the values lives in dedicated tags.type or tags.value fields.

tags = [{ type: CURRENCY, value: EUR }, { type: PLATFORM, value: STEAM }]

// Becomes this. Sub-objects field relations are lost.
tags.type = [CURRENCY, PLATFORM], tags.value = [EUR, STEAM]

The problem however is that once we set facetQuery to something like tags.type = CURRENCY, we still get all the other tag types like PLATFORM and COUNTRY.

How can we achieve what we want? What would be the best solution here and does Meilisearch even support this?

Would it make sense to split your types into different fields? I am thinking about something like the following to keep the relations together. When requesting facet distributions by using page 0, you'll get only the values associated with CURRENCY and no more STEAM value for it. You just need to ask about the facets distribution of tagsCurrency or tags.CURRENCY.

"tagsCurrency": "EUR"
"tagsPlatform": "STEAM"

// or even that way
"tags": {
  "CURRENCY": "EUR",
  "PLATFORM": "STEAM",
}

[rauldeheer]

Hey.

Thanks for your clear response and sorry that some parts of my issue were not clear to you. So basically, it's possible to do a facet on tags.value, because Meilisearch creates the following for this field:

tags.value = [EUR, STEAM, THE NETHERLANDS]

This makes sense to me and this also explains why we can already perform searches with filters inside these array of objects:

tags.value = EUR AND tags.value = STEAM

The only problem I'm left with now is that we have many more tags, but they're just not (yet) filterable on our website AND our products can have multiple country tags, for example. Obviously it would be neater if it just worked the way I expected it to work, but it's a search index so I guess your solution makes sense.

What would you do considering the fact we can have multiple tag types on one document? Would something like this work:

"currencyTags": ["EUR", "USD"],
"platformTags": ["STEAM"],
"countryTags": ["INDIA", "BANGLADESH"]

[Kerollmops]

What would you do considering the fact we can have multiple tag types on one document?

Yeah, classifying your filters will make the thing work. You'll be able to filter like the following and get the stats and facet distribution you want.

currencyTags = EUR AND countryTags = BANGLADESH

Thank you for using Meilisearch 🙂

[rauldeheer]

Great. Thanks! @Kerollmops