Combining embeddings with external metrics

Hi, my query is related to combining sentence embeddings and some external metrics. For the task of neural information retrieval, more specifically re-ranking, I have a few metrics such as page rank, etc. upto a max of 10 metrics (all numbers in a specific range) describing the quality of the webpage.

I have the sentence/document embeddings using sentence transformers and want to merge/concantenate these metrics to form a single document embedding vector for each document which can be compared with the query using some distance metric such as cosine to get the final ranking list. What would be the best strategy to go about it?

Would a simple concatenation of the metrics vector with the embeddings from sentence transformers and pooling work? Should I concatenate and pass them through an auto-encoder?
The metrics which I have carry a lot of weight in defining the webpage (quality), so I am not sure if simple concatenation with mean pooling will give expected results?

At the end, I would like to have/store a single document vector representing each document derived from text embeddings and external metrics.
I would really appreciate any help on this. Thanks.

No repository file, test, or entry point is named; the issue is a design question about combining sentence embeddings with external metrics for reranking. Start by reviewing the project’s embedding and retrieval documentation, then identify whether an existing API covers this use case. Done would require a documented, tested approach, but the issue does not define one.

Written by the indexing model from the issue text.