LLM Intent Classification

An Overview of Large Language Models (LLM) Intent Classification and Its Implementation in Parloa

Parloa's LLM Intent Classification enhances your bot's ability to understand user intents. This method complements the traditional Natural Language Understanding (NLU) approach, which relies on a trained speech model. Unlike traditional methods, LLM employs natural language descriptions, facilitating a more straightforward and effective process for building intent systems without the need for separate intent utterances.

This feature is available for Textchat 2 and Phone 2 platforms.

Activation Process

To begin using LLM Intent Classification, you must first activate it, as it is not enabled by default. Initiate this process by contacting our support team via email at [e-mail] or reaching out to your Customer Success Manager. Once activated, you can apply the feature to Start Conversation and State blocks in your bot's workflow.

Please note that employing LLM Intent Classification incurs additional charges. For detailed pricing information or to modify your contract to include this feature, please consult your Customer Success Manager.

LLM-based systems may exhibit some latency in intent recognition. Consider this factor when designing user interactions.

Use Case Benefits

• For new bots: Begin with LLM classification to minimize the need for numerous example utterances.

• For existing bots: Incorporate LLM classification to reduce reliance on speech models for intent recognition.

Enabling LLM Intent Classification on a Block

Step 1 – Selecting the Intent Detection Method

1. Click on the block you want to update (Start Conversation or State).

2. Navigate to the Intents tab -> Detect Intent by dropdown menu, as shown below:

If LLM is not enabled, the Detect Intent by menu will default to Utterances and you will not be able to select the Description method.

1. Select Description from the dropdown menu to set the intent classification method: An icon will appear on the block to confirm your selection:

   

Step 2 – Add Intent Descriptions

Provide an intent description in one of two ways –

• Click the edit icon next to the intent: The following displays:
• Alternatively, add intent descriptions by going to Speech Assets -> Intents:

An intent description is mandatory for functionality. Otherwise, an error icon will indicate a non-functional intent:

Crafting Effective Intent Descriptions

• Write descriptions in simple language, within a 500-character limit, that accurately reflect the intent's purpose.

• Ideally, descriptions should be between 70 to 125 words long.

Maintenance and Improvement

Frequently Asked Questions (FAQ's)

Can I continue using the Utterances method if I enable LLM Intent Classification?

Absolutely. You can continue to use the Utterances method even after signing up for LLM Intent Classification (Descriptions).

What happens if I enable the LLM descriptions but don't add any descriptions to the intents?

In such cases, your bot will trigger the fallback intent.

Am I still able to train my speech model with LLM Intent Classification enabled?

Yes, you can continue to train your speech model as usual. Utterances will remain visible and editable.

Can I turn off the LLM feature or deselect utterances entirely if I choose to?

Yes, you have the flexibility to disable the LLM feature or deselect utterances in certain blocks at any time.

How can I identify which of my blocks are powered by NLU or LLM?

You can easily distinguish them by the icon displayed on the Start Conversation or State block. For example:

• Utterances –
• Description –

Is it possible to enable LLM Intent Classification only for certain State blocks?

Yes, you can selectively apply LLM Intent Classification to specific State blocks.

Will the caller's experience change if I use LLM for intent classification?

No, the caller experience remains exactly the same when using LLM for intent classification.

Got more questions or want to share your feedback?

Please reach out to us at [e-mail].