In this article I provide a step by step guide for improving Dialogflow ES accuracy
Export your agent for versioning
First “freeze” your current agent version.
Remember that the accuracy only makes sense with respect to a given version of the agent.
Tip: You can also flatten your agent ZIP file and turn it into a CSV so it is easier to get a quick overview of the agent training phrases as you are calculating the accuracy. You can flatten your agent ZIP file using the Search tool.
Convert History JSON to CSV
Dialogflow ES does not provide an easy way to download or export agent conversations.
But you can convert your history JSON to CSV and upload it to Airtable so you can calculate the accuracy.
Annotate user utterances
Use the formula I have explained before to annotate the user utterances in Airtable.
Classify the user utterance into one of the four categories – True Positive, True Negative, False Positive and False Negative.
Calculate the accuracy
Use the confusion matrix formula to calculate the accuracy of your bot. If you used Airtable to annotate user utterances, you can calculate this quickly.
Update your bot
Now make changes to your bot.
You can use automated conversation tests to make sure you don’t break your bot’s behavior as you are making your updates.
Once you are finished with your updates, go back to step 1 and repeat the process.