Why use Intent Training?
A good intent structure entails a good AI model that understands the most frequent intents and ensures your virtual agent takes the automation potential to the fullest. But what is a good intent structure? A good intent structure should reflect the reality of your support query frequency. Meaning the most frequent intents should have the most expressions.
When to use it?
Before launching the bot, aim for 50 expressions per intent as the baseline and 150-200 for the most frequent ones.
In addition, before or after launching the bot, if you notice that your most frequent intents have fewer expressions than the other less frequent ones.
Training Expressions to an Intent
To train the intents that need to be strengthened:
- Navigate to Training Center and click Intent Training
- Select the intent you want to add more expressions to
- Use the search bar if you've got a long list of intents
- You'll see two tabs on the right:
- Take a brief look at Current Expressions to get an idea of what expressions you should be looking for in AI Suggested Messages. Then, go to AI Suggested Messages.
- In AI Suggested Messages, select the messages that should be trained to the intent, and click Train.
This view shows the expressions suggested by the AI model that could be added to the intent. All these actions can be done in bulk or one by one.
To add expressions, simply type the expressions you'd like to add in the bar on top then hit enter or click Add Expression.
Move expressions to another intent
If you spot any expression that should be trained to another intent, you can do so by clicking the forwarding arrow of an expression, or selecting multiple ones and clicking the forwarding arrow on top.
If you spot any expression that shouldn't have been trained to this intent, you can untrain it by clicking the X icon of an expression, or selecting multiple ones and clicking the x at the top.
The untrained expressions will be returned to where they are from originally - meaning if it was trained from another intent, it will go back to that intent. If it was untrained previously, it will remain as an untrained expression.
AI Suggested Messages
This view shows the expressions suggested by the AI model that could be added to the intent. The untrained expressions you see here can be adjusted using the Confidence Range slider.
This article covers:
The default is set to between 50% and 70% as this is usually the sweet spot that provides the most value when training the bot. We suggest setting the maximum to right below the confidence threshold of your bot to get the best performance. The confidence threshold of your bot can be found in Settings > General > Confidence Threshold for default messages.
Train to another Intent
If you spot any expressions that should go to another intent in this view, you can also train them to the suitable intent by clicking the triangle next to Train and then selecting Train to another intent.
Train to a new Intent
If you can't find a suitable Intent for an untrained expression, you can create a new Intent and train the expression there simultaneously. To do so, click the triangle next to Train then select Train to a new intent.