What if the retrieved ddl is not related to the user question?

[Jackie-shi]

When do 'ask', the process here is to use vector database to get the most related DDL. But there is a situation that all of the ddl has nothing to do with the user question, which means that the final SQL Vana generates will definitely WRONG!

So I wonder if there is any mechanism or post-process to check or evaluate the retrieved ddl.🤔

THX!

[Jackie-shi]

What's more, when you do embedding, the raw ddl is not a natural language so the problem between ddl and user question is that they cannot be aligned in semantic space. It may make the result inaccurate!

[zainhoda]

Extra information is not usually detrimental to the SQL output, particularly if you have a lot of question-SQL pairs