I am curious to apply the embedding example to structured (relational) data for store items. The example shown in the course worked with structured data. Many real-life examples may have to deal with relational data. We can imagine that there could be multiple attributes of an item within the item table. Is there a standard or best practice to do that?
My idea on this is to take the structures data and create sentences with it. those sentences will be inputed to the LLM.
As simple as “#{word1} + #{word2} + …” or “When #{word1} happends #{word2}”.
I may guess that with those sentences you can make a custom LLM training specific for your domain.