Generating vector embeddings from multiple attributes

Question

To implement fuzzy matching between items, I'd like to generate vector embedding from items with multiple attributes. Those attributes include:

1. text fields with phrases
2. a list of email addresses (a strong indicator of similarity)
3. several numeric properties

Most examples only show how to create an embedding from an unstructured block of text. My understanding is two techniques would need to be applied:

1. generate partial embeddings for each of the attributes
2. combine all the embeddings

How can partial embeddings be generated for the non-text (numeric and email list) attributes? Are there any techniques or libraries for combining embeddings with weights? Or is this a wrong-headed approach to similarity matching for this kind of data?