The intent section is where you will create and include the knowledge and training for your agent, which will be used to classify the user's intent.

Your agent will use this list of utterances to classify user inputs.

Every example you will enter should represent a sentence callers may use to attain a specific answer or action. Add as many unique examples as possible to make sure your agent is efficiently trained and able to understand the different ways people are referring to one query.

You can either create a new intent-based on your use case or import an intent from an already existing agent.

Creating an Intent

Here's how to create your own intent!

  1. In the ToolBar on the left, click on Properties > Intents.

  2. In drawer that will open on the right, and select ‘Create Intent' in the top right.

  3. Name your intent - This should be something that tells you more about its content and purpose.

  4. Insert all the relevant training in the ‘User Expressions’. These user expressions or training phrases denote all the different ways a user can signify the intent.

You can also create a new intent in the Classification node by pressing ‘Add Intent’ clicking the drop-down, and then ‘Create Intent’.

How to train your agent

Train your agent primarily based on keywords

You can add full sentences but it is not required and should be minimal. When adding user expressions, it is important to keep the essence of the intent and leave out all the meaningless parts of a sentence.

E.g. "I want to request a loan" - Add "Request loan" to the training set

Be creative

Some people refer to "scheduling with someone" as a "meeting", others might call it an "appointment". Try to use various different wording and phrasing to refer to the same thing.

E.g. Meeting: Appointment, Interview, Consultation, Date, Meet, etc.

Once the agent is live and interacts with users, it is helpful to add more training based on realtime conversations with the users.

How to use intents

Heres a visualisation of how you can create intents. The example use case used here is that of Credit card balance.

Step 1

After the initial greeting, for example, collect the input of the user in a Collect Input or Listen node. Under the parameter to be collected, add a TOPIC parameter connected to a sys.any entity.

Step 2

Next, add a Classification node and select the same TOPIC parameter under "Classification parameter".

Now it's time to add intents.

In the Toolbar on the left, go to Properties and select Intents. If you haven't already, click on "Create intent". For the sake of the example, name it "balance" and train it with the keywords and synonyms around that topic.

E.g. "What's my current account balance", "How much money do I have on my account", "Am I broke?", etc.

You can check in the search bar for a specific sentence you added prior. Just type in the sentence you are looking for and it will pop up in the user expressions.

Step 3

From here, you can either add more intents based on your use case or go ahead and build out your flow.

For more information on how to create your first conversational flow, click here.

Export an Intent

You can export an intent by clicking on the little export symbol on the right of the intent. This will export a CSV file containing all intent expressions.

Import an Intent

  1. When in the intent window, click 'Import Intent' on the top right.

  2. The AI Studio will prompt you to select the relevant file from your device. The file format should be CSV. The name of your file will be the name of the intent.

  3. The format of the .csv file should be as follows:-

Please include the "Name" and "User Expression" headers under which the name of your intent and the training phases are allocated respectively. Separate your training phases with no spaces and singular commas. Please note you can only import one intent per .csv file. Which means if you have multiple intents its better to store them as separate files.

5. If needed, you can always make changes to the intent. Don't forget to hit "save" once you're done adding new expressions.

Minimalism is key - This applies to the flows you create, number of intents and training data. Keep your flows straightforward and short whilst also incorporating fallback options. We recommend having less than 20 intents in total to improve classification scores. The more intents you have the more ambiguity the agent has to deal with. Keep your training set for each intent up to ten unique synonyms for best results. Avoid including whole phrases like “ I want to rent a bike” rather stick to the topic and only include “rent a bike”. Try your best to not include the same word within the training sets of different intents.

Last updated