Add more training phrases to CSV file for Dialogflow CX

[Yannick_SG]

Hey community,

our company want's to use the data storage feature of "Search and Conversation AI" to create a FAQ bot using Dialogflow CX.

I want to provide our customer support with a Google Sheet file, which they can fill out to provide common question variants to an answer. But how do I have to format the csv file, so  I can give several training lines?

Currently my setup is:
ID | question | answer

I tried it like this before, but then it stopped working:

ID | question 1 | question 2 | question 3 | answer

Or should each question have it's own row and copy the answer?

Before I had a "pattern" - menu here, where I was able to manually add types to a column, but I don't see it at the moment.

[cdoran]

I believe each should have its own row.  The example given in the docs is:

"answer","question"
"42","What is the meaning of life?"

https://cloud.google.com/generative-ai-app-builder/docs/agent-data-store#structured_data_store

[jasonparker]

Yes, you should have the data in "long" format, one question and answer per row. It's OK if the answer is duplicated, with the semantic search workload it's looking for a match on the question and then will use whatever answer is associated with the question that is most relevant.

[Poala_Tenorio · Accepted solution]

For Dialogflow CX's Search and Conversation AI data storage, the preferred format is to have each question in a separate row with the corresponding answer. The format would look like this:

ID | question | answer
1 | What is your product? | Our product is XYZ.
2 | How can I contact support? | You can contact support by...

Each row represents a training example with a specific question and its corresponding answer. This format allows the model to learn patterns and variations in user queries.

If you want to provide multiple variations for the same question, you can create multiple rows for that question, each with the same answer:

ID | question | answer
1 | What is your product? | Our product is XYZ.
2 | Tell me about your product | Our product is XYZ.
3 | What does your company offer? | Our product is XYZ.

This way, you ensure that the model understands different ways users might ask the same question.

Regarding the "pattern" or "menu" feature you mentioned, it's possible that the interface has been updated or changed. I recommend checking the latest documentation or user interface guides for Dialogflow CX to see if there have been any changes to the way you can specify patterns or menu options for intent recognition. You may also want to explore the options available in the training phrases and entity recognition settings within the Dialogflow CX console.

[Yannick_SG]

Thank you, I tried it like this and it works fine.