The NLP dashboard (and the tab 'improve') give you an overview of the quality of your NLP model. It automatically detects any overlap between intents, which is especially useful if you are building a larger NLP model, or if multiple team members work on it.

NLP Dashboard

The NLP dashboard lets you know how well you are doing for each language. Based on how you have trained your NLP, it gives you an overall model score, ranging from 0% to 100%. The overall model score is calculated based on the amount of intents with too few expressions and the amount of intents with misclassified expressions.

  • On the left side of the dashboard you will find a general summary of your model

  • Below you can see your training history which shows how you're doing so far and whether or not you are improving over time

  • On the right side, you find an overview of all the intents with know issues. The blue number between brackets indicates how your score has changed since your last training

Training status

When training an NLP model, there are four possible statuses:

  • Ready: This means that your NLP model is active and ready to make predictions.

  • Unloaded: This means that the model is inactive and has not made any predictions or has been trained in the last three months.

  • Cancelled: This means that the training process has been cancelled. This status can only be set by the Chatlayer internal team.

  • Failed: This means that the training process has failed and the model is not available for use.

Not enough expressions

A first metric our platform measures is the amount of expressions. To train your model as best as possible, we advise to create at least 20 expressions per intent. Ideally you'd have around 50 expressions per intent. The more, the better, and the more accurate your bot will be able to interact with the user.

Misclassified expressions

Just like us humans, no AI or NLP system is absolutely perfect. And because our brain works different from the Chatlayer.ai NLP, it is hard to predict where mistakes can and might happen.

To guarantee the best possible outcome for your bot, we do an advanced analysis of all the expressions. This recognizes the expressions that the NLP might have a hard time with. For example, it might have difficulties differentiating between sentences like: β€œI see on my bill that I have the wrong subscription” and β€œMy bill seems wrong if I look at my subscription”. Both sentences mean a different thing. So the second metric we take into account is the number of intents that have a risk for misclassified expressions like these.

In the intent list on the right side of the NLP Dashboard, you can click on the wrench icon to improve your set of expressions and thus to better differentiate between intents.

Last updated