I have a dataset that contains columns in the form of [‘title’, ‘subtitle’, ‘content’]. When creating embeddings, I only send content to embed, do you think this approach is correct? When adding data to the vector database, I use the title and subtitle columns to indexing.