feat: MetaField Ranker

[domenicocinque]

Related Issues

• fixes Ranker based on custom meta field #6054

Proposed Changes:

• proposal: meta field ranker #6141

Why:

To allow users to rank documents by a relevant metadata field after having used a retriever.

How can it be used:

from haystack.preview.components.rankers.meta_field import MetaFieldRanker
from haystack.preview.dataclasses import Document 

# Documents coming from a retriever 
documents = [
    Document(content="Product 1", meta={"rating": 1.3}, score=0.3),
    Document(content="Product 2", meta={"rating": 0.7}, score=0.4),
    Document(content="Product 3", meta={"rating": 2.1}, score=0.6),
]

ranker = MetaFieldRanker(
    metadata_field="rating",
    ranking_mode="reciprocal_rank_fusion", 
    weight=0.5
)

sorted_documents = ranker.run(query="", documents=documents)
print(sorted_documents)

The example shows how the component can be used to rank documents by combining a meta field of choice ("rating" in this case) and the score of the retriever.

How did you test it?

Tested locally and with unit tests

Notes for the reviewer

The implementation is based on the Recentness Ranker. However I made some changes such as renaming the "score" ranking method to "linear_score" to make it more specific. Moreover I separated the logic that reranks the results in another function, in order to make it possible to inherit from this class for the implementation of a Recentness Ranker in Haystack 2.0

Checklist

• I have read the contributors guidelines and the code of conduct
• I have updated the related issue with new insights and changes
• I added unit tests and updated the docstrings
• I've used one of the conventional commit types for my PR title: fix:, feat:, build:, chore:, ci:, docs:, style:, refactor:, perf:, test:.
• I documented my code
• I ran pre-commit hooks and fixed any issue

[vblagoje]

Hey @domenicocinque , thanks for opening this PR. Would you please provide a bit more context about using this component? See, for example How it can be used section of #6199 It'll help greatly not only me but also @dfokina who will help us with the docs.

For the release note CI failure, you need to add a release note with reno tool. See https://github.com/deepset-ai/haystack/blob/main/CONTRIBUTING.md#release-notes for more details.

[domenicocinque]

Hi @vblagoje, thanks for the fast response. I added the How it can be used section in the PR description and the release notes in the code. Please let me know if it needs to be improved

[vblagoje]

@domenicocinque Your explanation helped me a lot to understand the use case of this component. However, what I'm not 100% sure about is whether this component should be integrated into our core packages or included as a valuable community integration. Let me consult internally and we'll get back to you soon.

[vblagoje]

@domenicocinque looks quite solid already. I left one comment that I think should improve code readability and performance as well

[vblagoje]

@domenicocinque great work, thank you. One last request - let's add a unit test for the non-happy path in linear_score mode when the score is invalid. Tests those logs in the caplog are captured. Also, in init, let's raise ValueError instead of ComponentError. ComponentErrors are mostly reserved for the run method to signal the invalid component state preventing execution.

[vblagoje]

@dfokina I don't expect any more changes for this PR after @domenicocinque's next commit. Please have a look at it after he commits his last change and make any pydoc corrections 🙏

[vblagoje]

Thanks for the update @domenicocinque and for this overall valuable contribution! @dfokina have a pass now, make any needed changes, and we are ready to 🚢

[dfokina]

All done from my side too! 🚀