@yushengliao we could add a score to the Content class, but it will only work for Contents retuned by EmbeddingStoreContentRetriever, other retrievers (including custom ones) won't have it

@langchain4j
Could we consider returning the embedding_id within the Content, so that, in conjunction with the returned score and the existing metadata and text, it would provide more flexibility for handling various business scenarios?

@fmx0717 we probably could. You can also store the ID in the metadata of TextSegment though.

I guess we can add a Map<String, Object> metadata field to dev.langchain4j.rag.content.Content and store score and embedding id there. WDYT?

@langchain4j
Of course, it's a great idea.

@fmx0717 would you like to open a PR?

@langchain4j
Thank you for your trust and the opportunity. I'm very interested in the project and will definitely consider contributing in the future when I have more time. However, I am currently quite busy with other commitments. Thanks again for understanding!

@langchain4j

Perhaps I would like to complete this PR. Could you please advise on how to implement it effectively?

1.Add EmbeddingScore and ReRankScore member variables in the Content class?
2.Or add key-value pairs for these scores in Content.TextSegment.Metadata?

@yushengliao I would probably introduce a Map<String, Object> metadata field in the dev.langchain4j.rag.content.Content and store score(s) and other things (e.g. embedding id) there. WDYT?

We also need to make sure that ContentAggregator works properly after these changes and does not take this metadata map into account when comparing two Contents (e.g., for RRF)