Intent

This is where you define and manage the primary actions or goals users might express when interacting with your chatbot. Simply put, an intent captures what the user wants to do. By setting up and refining intents here, you help your chatbot understand and respond to user messages more effectively.

The vector classifier model is the main methodology, enhancing the chatbot's intent recognition capabilities. Additionally, users have the flexibility to integrate their custom NLP models by configuring settings in the chatbot's settings page. This empowers users to tailor intent recognition to their specific needs, ensuring a more personalized and accurate chatbot experience.

Creating an Intent

  1. Navigate to the Create Intent area.

  2. Enter a unique name in the Intent Name field.

  3. Add new Samples:

    • Samples are example user inputs that depict various expressions of the intent.

    • By providing samples, you enhance the chatbot's ability to accurately identify the intent, even with different phrasings.

    • You can add up to 100 samples for each intent.

  4. Click Create to finalize the intent creation.

Editing an Intent

  1. Choose the intent you wish to edit from the list on the left side.

  2. Modify the necessary fields, following the guidelines provided in the Create Intent section.

  3. After making all desired changes, click Update.

Deleting an Intent

  1. Select the intent you wish to delete from the list on the left.

  2. Click on the trash icon.

  3. Confirm the deletion by clicking Delete.

Searching for an Intent

  1. To search, ensure your query begins with the correct starting characters.

  2. Type in the name of the intent in the search bar and either press Enter or click Search.

  3. To revert back to the intents page, clear the input field and either press Enter or click Search.

  • You can only search using the intent name.

  • The entities list is organized by the Last Modified Date.

LLM Model Classifier

Recommendation Tips

  • When adding samples, it is recommended to start with 3 samples that clearly convey the same meaning as the intent.

  • If the bot does not classify accurately, gradually increase the number of samples to 5, then 7, as needed.

  • When adding or creating intents, avoid using similar samples across different intents.

Naming Limitations

  1. Intent names must be between 2 to 100 characters.

  2. Intent names must be in lowercase English letters only.

  3. If spacing is needed, use an underscore (_) instead.

  4. Intent names must reflect the meaning of the training samples, as the LLM model interprets intents primarily based on their names.

    • Example:

      • Old intent name: ChangeBeneficiary

      • Updated intent name: change_beneficiary

Sample Addition Limitations

  1. Each intent must have at least 2 samples and no more than 100 samples.

  2. Each sample must contain 2 to 500 characters.

  3. Each sample must convey the same meaning as the intent name to avoid misinterpretation.

  4. Samples can include wrong typos, as long as the intended meaning remains consistent with the intent name.

  • Example:

    • Intent name: health_insurance_inquiry

      • English

        • I want to ask about health insurance.

        • Do you have any health insurance plans?

        • Can you give me details about your health insurance?

        • pleas tell me about helth insuranc (intentional typos)

      • Thai

        • มีแผนประกันสุขภาพอะไรบ้าง

        • ขอรายละเอียดของประกันสุขภาพหน่อยค่ะ

        • ขอทราบข้อมูลประกันสุขภาพ

        • ประกันสุขภาพของคุณครอบคลุมอะไรบ้าง

Creating a LLM Intent

  1. Importing a CSV file (recommended when there are multiple intents and samples to be trained at once)

    • Go to Chatbot > Intent Click Import Intent

    • Click Download CSV Template to download the format for importing intents

    • You will get a template as shown in the example. You can format the file to be used on the bot platform as follows:

      • Enter the intent name under the header _name (column A)

      • Enter the sample phrases under the header _value (column B)

    • Select the prepared .csv file from your device to upload to the bot platform

    • Once the upload is complete, a confirmation message will be displayed as shown

    • Once the intent is successfully created, its details will be displayed. You can verify this by checking the Last Modify Date, which will show the timestamp of when the intent was created.

  2. Manual creation

    • Type the desired intent name in the Enter intent name field

    • Enter a sample phrase in the input field

    • Click the ‘+’ button to add the sample

    • After adding all the desired samples, click Create

    • Once you click Create, a confirmation message will appear on the screen as shown

    • Once the intent is successfully created, its details will be displayed. You can verify this by checking the Last Modify Date, which will show the timestamp of when the intent was created.

Update a LLM Intent

  • Type the intent name you want to update in the search intent name field

  • Select the intent you want to update

  • Make the desired edits

    • You can delete a sample by clicking the ‘x’

    • You can edit an existing sample by clicking on it and modifying as needed

    • You can add new samples by following the steps in the How to create an intent section

  • After making the necessary changes, click Update

  • Once you click Update, a confirmation message will appear on the screen

  • Once the intent is successfully updated, you can verify it by checking the Last Modify Date, which will show the timestamp of the latest update.

Training a LLM Intent

  1. Once you have created and updated the intents as desired, the next step is to click the ‘Train’ button to start training the intents.

  2. You will see the ‘Training…’ status on the button, indicating that the bot is currently training. The training process typically takes around 45–60 minutes.

  3. When the training is complete, the ‘Training…’ button will revert back to ‘Train’. You can also confirm that the training has finished by checking the Last train timestamp, which should reflect the current time after training completion.

FAQ

Q1: How to set up unknown_intent (fail flow) for LLM?

A1: Since the LLM will try to classify messages into existing intents on the platform first, you need to create an additional intent to filter out messages that should go to the fail flow. These messages will be routed to this intent and then mapped to trigger the fail_flow action

A good unknown_intent must have samples where each sample has a completely different meaning from one another. (Samples must not point in the same direction or imply any other existing intents in the bot platform.)

PreviousChatbotNextEntity

Last updated

Was this helpful?