Is it possible to get the search result score?

[FourierTransformer]

I'm curious if I can get the internal ranking/score when performing a search.

[Kerollmops]

Hey @FourierTransformer,

Yeah, this is planed and is the 6th step of meilisearch/meilisearch#692.

[FourierTransformer]

That sounds great, thanks! Meilisearch helped make adding search to my site really easy, and the search results are really good! Thanks for all the work the team is putting in!

[raine]

Would this be part of a solution to being able to filter out only remotely accurate search results from otherwise accurate?

For example, I have a case where 6 of the results contain the search keywords in the title field, but there's also 3 other results that have them somewhere in the fields.

I would imagine you could use the disparity in score to filter out the other three.

[tpayet]

Hi @raine, you might want to have a look into the ranking system. You should be able to filter out the three results that got the query keywords not in title by removing those fields in the searchable attributes: https://docs.meilisearch.com/guides/main_concepts/relevancy.html#changing-the-order-of-the-attributes

[JonathonWebster]

Hi @Kerollmops,

Would this return a numerical value for each ranking rule? I'm hoping to use the numerical values to create a custom rank based on the existing rules. This would allow me to have a less linear ranking rule priority (if i say typos is the number 1 priority, and number of words matching 2nd priority, something with only 1 word missing might still be better that something with 5 typos).

I haven't started playing with this directly yet, but after reviewing the documentation, it seems like I would need something like this in order to optimize my ecommerce product feed. A more likely example would be needing to weigh estimated revenue and overall relevance - in which case I would still need to design a consolidated 'relevance' score.

Lastly, in the first example I mentioned, do the ranking rules truly work like this, where if I choose typos ranked as #1, it will give me all results with fewest typos, regardless of how well the other rules perform?

Thanks for your help!

[MihaiNueleanu]

Hi ! It's been a while since the question was originally asked. Is this feature still planned for development?

[curquiza]

Hello, the feature suggestions are now handled on the product repo :) here is the dedicated discussion about it, I've just opened it since it looks like it has not been created yet #379
It has not been prioritized and is not planned since not a lot of users requested it so far compared to other features.

If you have time, you can share your need and your usecase. It would be definitely helpful to know what you need and to maybe prioritize it 🙂

[ar-siddiqui]

I would like this feature to get search results of different indexes and then merge and sort them using relevancy score.

[Nyholm]

Yeah. That would be great.

Do you have a name to share?

We use a custom TF-IDF and a plain database search.

[Snarik]

Im just going to chime in and say I would also very much like a normalized score with the hits.

[Kerollmops]

Hey everyone 👋,

We are looking into this feature as we are working on returning the ranking rules scores in a future version of the engine.
Would the ranking rule score be enough/adapted for your use cases?

The ranking rule score corresponds to the score each ranking rule will return in the rankingRules settings order.
I understand that it can be hard to order an array of scores especially if you don't know if a certain element must be ordered in descending or ascending order, we will take that into account to expose a better API.
Here are some settings examples and the corresponding score that could be returned for each document:

{
  "rankingRules": [
    "words",
    "typo",
    "name:desc",
  ]
}

[
  {
    "document": {
      "title": "I am the best document",
      "name": "kero"
    },
    "score": [
      3, // matches 3 words of the query
      0, // no typos
      "kero" // we can't compute a normilized score for a string so we put it here.
    ]
  },
  {
    "document": {
      "title": "I am the second document",
      "name": "pierro"
    },
    "score": [
      3, // matches 3 words of the query
      1, // one typos, that's what moved this document down
      "pierro"
    ]
  }
]

About the score ordering we could return an array explaining the order of each ranking rule and their names:

[
  { "name": "words", "ordering": "desc" },
  { "name": "typos", "ordering": "desc" },
  { "name": "name:desc", "ordering": "desc" }
]

[inspire22]

That would be very helpful, thanks :) A simplified "rank" score (just the other ones summed?) would match what mysql's fulltext search provides, but we could always calculate that manually from the full list of sub-scores. I'm using meilisearch to detect plagarism using fuzzy searches, then for high-scoring results doing a Levenshtein distance or ML driven more complicated final ranking from that subset (if needed, the score might actually be enough). Meilisearch seems awesome! Are there beta branches followable to try stuff like this out early?

[gmourier]

Hello @inspire22 👋

A simplified "rank" score (just the other ones summed?) would match what mysql's fulltext search provides, but we could always calculate that manually from the full list of sub-scores.

Do you mean having a score between 0 and 1 ?

Are there beta branches followable to try stuff like this out early?

Not yet! But we will work on it in Q2/Q3 😇 We will keep everyone updated whenever we move on that

[inspire22]

Mysql's fulltext score seems to be an arbitrary number, I'm not sure if there's a maximum. For me it was usually between 0 and 70 or so. Whatever works. It can be nice to know when the top result (or top 3) is 50x stronger of a match than the next match, for example.

[macraig]

Hello everyone! 👋

Quick update on our side, we've started working on a solution to return ranking score details on search.

You can follow the updates in #594

[macraig]

Update: Scoring details will be available as an experimental feature in Meilisearch v1.3 (released July 31st)

You can find the details in #674

[Snarik]

extremely exciting!

[macraig]

Hello everyone 👋

We have just released the first RC (release candidate) of Meilisearch containing this new feature!
You can test it by using

• the release assets
• the Meilisearch Docker image

docker run -it --rm -p 7700:7700 -v $(pwd)/meili_data:/meili_data getmeili/meilisearch:v1.3.0-rc.0

You are welcome to leave your feedback in this discussion.

If you encounter any bugs, please report them here.
Thanks in advance for your help and your involvement in Meilisearch ❤️

🎉 The official and stable release containing this change will be available on July 31st, 2023

⚠️ RC (release candidates) are not recommended for production

[macraig]

Hey folks 👋

v1.3 has been released! 🦁 You can now get global and detailed ranking scores for your documents ✨

Note: Document ranking score details are considered Experimental, you can find the usage instructions in #674

📚 https://www.meilisearch.com/docs/learn/core_concepts/relevancy#ranking-score
📚 https://www.meilisearch.com/docs/learn/experimental/ranking_rule_score_details