urgent help needed with DF intents

[Shlomi Ben Ami]

Hi,

We have built an agent to identify a few intents from our customers.

We do not use it in a chatbot - we use it for emails our customers are sending in. We are pushing the email sentence by sentence through DF and get the confidence for each sentence.

The two intents are refund requests from customers and cancellation requests for their subscriptions.

The two intents are well trained with the necessary phrasing. And basically it works pretty well.

But we have very strange results sometimes. For example, whenever there is a name mentioned (the customer wrote his first name in the signature of the email), and it is passed through DF, it detects it as one of the mentioned intents above, with high confidence!

It kills me as I do NOT manage to understand why! If I add that name into the default fallback, then the next time it will not identify it as one of the real intents - OK, so I know how to fix a certain case.

But then the next name again, gets identified as a real intent.

What am I doing wrong?

[Monica (Google Cloud Platform)]

Hello,

Please open a Private issue with the Dialogflow team. The Google Groups are reserved for general questions on Google Cloud Platform-end products. 

You should provide the agent zip, the reproduction steps and screenshots that illustrates the issue. 

Also, please run the Agent Validation if you haven't done it already, to get suggestions on improving the quality and performance of your agent.
Thank you!