Suggestion for Ranking texts based on premise

I have a premise text and multiple following texts.
I want to rank these following texts based on the premise. These following texts will be short like 3 or 4 words. I want to rank them in a way that the comlete sentence makes sense.

The data is related to a chat conversation.

I have couple of things in mind -

  • train a sentence transformer and use a embeddings to calculate similarity
  • train a sentence pair cross-encoder and use the ranking to rank the following texts.

Is there something else i can play with?