[Feature Request]: Hope to add rerank models to the project

[xxxl123]

Is there an existing issue for the same feature request?

• [X] I have checked the existing issues.

Is your feature request related to a problem?

No response

Describe the feature you'd like

I think RAG+re-rank is a great combination that can improve recall accuracy.

Describe implementation you've considered

No response

Documentation, adoption, use case

No response

Additional information

No response

[Nuclear6]

I see that there is a rerank module in the project architecture diagram. Does the module they provide fail to meet the requirements? Or some other problem?

[xxxl123]

I see that there is a rerank module in the project architecture diagram. Does the module they provide fail to meet the requirements? Or some other problem?

I asked the developers, and they haven't added the rerank model yet.

[Nuclear6]

What you mean is that the rerank here is only based on some strategies to achieve the effect of rerank, and does not involve reranking LLM, right?

[xxxl123]

There should be a rearranging strategy, but the rerank model was not used.

[Nuclear6]

So I have a question, what rerank models are available here? Have you ever compared the results?

[KevinHuSh]

Weighted keywords and embedding similarity is RAGFlow's reranking model right now. And we're working on replacing the embedding similarity with other so called rerank models like BGE reranker and others.

[xxxl123]

I think the accuracy of the recall is already very accurate now, and I hope it can be further improved. It shouldn't be worse after adding it.