Getting Started with the Intent Model

Intent recognition is a smart tool in AI and machine learning that helps you teach your chatbot to understand complex and diverse questions from your users better. The term "intent" also gives you a clue about what the tool is trying to figure out in these questions: why the user is asking them in the first place.

Users might be looking for general or specific information, trying to become a customer, checking a status, or having trouble with a particular process. To make your chatbot good at this recognition, you need to teach it by using your data.

Generative AI can assist you with this during some steps of the intent modeling.

An intent in Conversational AI Cloud consists of the following elements:

• Intent name

• Description

• Utterances

• Test phrases

• Articles

Intent Name & Description

When you create a new intent, you will need to provide a name and description. The intent name and description will help you to identify what the intent is used for. The description provided here will also help the Generative AI feature to produce some example phrases later on for your utterances and test phrases.

Utterances

Utterances refer to the various ways or expressions that users may use to communicate a specific intention or request. Utterances are the user's input, which can take the form of spoken or written sentences, phrases, or questions. These utterances are examples of how users interact with a system, and they are used to train the intent model to recognize and understand different variations of a particular intent.

Conversational AI Cloud uses Generative AI, so you can easily can generate utterances for your intent with a simple click of a button. It is recommended to have at least 10 utterances per intent.

Enter the number of test phrases that you want to generate:

These diverse utterances help the intent model learn to identify the common entities and keywords associated with the "Order Pizza" intent, enabling it to correctly recognize and respond to similar user requests, even if they are phrased differently.

Click "Confirm" to add the generated utterances to the intent.

Test Phrases

Test phrases are a set of sample queries that are used to evaluate how well the model can correctly recognize and classify user intentions. They serve as a way to assess the accuracy and effectiveness of the intent recognition process. Using the example of the "Order Pizza" intent, let's explain test phrases:

We've just added various utterances for ordering a pizza. Now, we want to test how well the model can identify the "Order Pizza" intent from new, unseen user inputs. To do this, lets generate a set of test phrases. Note that they should be different questions than the utterances we just added.

Click "Confirm" to add the test phrases to the intent.

Articles

Now we need to add an article for our intent to know which answer to provide to our end user.

Click the "Link Article" button:

Search the answer that you want to use and click "link" at the top of the article:

Train the Intent Model

Your intent now contains all of the elements needed to recognize your end user's questions about ordering a pizza. Now all we need to do is train the model! Go back to the overview and notice the "Training required" reminder next to your newly added intent.

Click the "Train" button at the top of your intent model.

You should notice the model score improve after adding and training a new intent. Now you can test it in the test center to see how it works!