Our experimental initiative to prototype a bot-ready information solution using Google’s Dialogflow

This post is part of a series. For more information and links to other posts in the series, see the “Building bot-ready knowledge bases” home page.

The Dialogflow console has links to the training and validation (troubleshooting, debugging) features.

Training

The Training tab displays statistics about how often requests were handled by the various defined intents, and when was the last request date. You can also see a log of all requests handled by a given intent. No information is provided for requests handled by the Knowledge Base.

We did not make much use of the Training feature.

Validation (Troubleshooting, Debugging)

Validation feature in Dialogflow

The Validation tab displays a list of issues Dialogflow has found during the time you define and train your bot. We used this information to improve the training for some of our defined intents.

The image below shows that Dialogflow has identified that one of our intents does not have enough training phrases.

Debugging and fixing an intent issue

When we displayed the list of phrases in the intent, we saw that we had only three.

Adding more phrases and retraining our bot fixed the issue.

Diagnostic information provided by Dialogflow

As bot builders, we added source information to some of the bot’s responses. For example, in the response below it is useful to know that the response came from the “Produce: Overview” Knowledge file.

Clicking “DIAGNOSTIC INFO” in the image above opens the window below.

Information in the JSON file above tells you which intents were considered by the bot.

What’s next?

In our 7th post in this series we talk about publishing GROCERYbot as a web demo.