Intents Migration Guide

Migrate your intent model from V1 to V2, or migrate from entity based recognition.

Release 4.8 of Conversational AI Cloud brings a range of changes to its intent recognition, recognition dashboards, and the overall Conversational AI Cloud CMS. In this article, we will outline the necessary steps for migrating from different scenarios to the new intent recognition model in Conversational AI Cloud. It is worth noting that the use of intent classification is optional, as Conversational AI Cloud offers a multi-engine NLU and customers can choose which parts of the NLU pipeline to use. Let's start by examining the major changes and new features that you need to be aware of.

What’s New

With the 4.8 release, Conversational AI Cloud proudly introduced various new features, such as:

• Generative AI capabilities that enable the generation of intent recognition models, utterances, and test phrases

• Direct links between intents and articles that give customers explicit control over the articles matched under an intent
• Direct links between intents, entities, and articles that allow customers to set the conditions under which articles are matched under an intent

• A new recognition analysis dashboard, which is only available if your project has a trained intent model

• Additional insights in the test center that display the exact recognition, context, and state of the conversation that resulted in the answer you received

In addition, the following terms have been renamed to align with the direction of Conversational AI Cloud:

• Training phrases are now referred to as "Utterances"

• In the dashboards:
• The “Recognition Analysis” dashboard is now called the “Recognition Analysis” dashboard, if you are not using intent classification for your project the “Recognition Analysis” dashboard will contain the exact same data you were
  used to seeing in the “Recognition Analysis” dashboard.

• "% Questions Answered" is now known as "Recognition Rate"

• "Recognition Rate" is now called "Recognition Quality"

Migrating From Entity Based Recognition

If you’re currently using Conversational AI Cloud’s entity based recognition, you don't need to worry about making any changes to your existing content! Simply choose a conversational use case that you'd like to cover with intent classification and follow
these steps:

• Create your first intent

  • Be sure to add enough utterances and test phrases, preferably questions your customers asked as shown in your recognition analysis dashboard

• Select the article that you want to be returned whenever the intent is matched

• Train your intent model

• Publish the intent model

After publishing, Conversational AI Cloud will automatically classify the intent for your customers’ questions. If it can classify the intent, it will match the article that you’ve defined. If there is no match, your entity recognition model will take over, providing you with the same high-quality recognition you had in your live environment before publishing your intent model!

Migrating From Intents v1

If you already had a running intent classification model in your live environment, you still have access to the interface and options that were available prior to release 4.8. To take advantage of the new capabilities introduced with release 4.8, follow these steps:

• Navigate to your project in the Conversational AI Cloud CMS

• Take note of each intent and its implicitly attached article

• Navigate to your project in the Conversational AI Cloud admin portal

• Under Settings, switch the intent model version to V2

Ensure that you read the following steps carefully and have adequate time to complete the migration process, as switching the intent model version cannot be undone without the assistance of the Conversational AI Cloud support team.

• For each intent, go to the Conversational AI Cloud CMS and:
• Open the intent Link the article that you want to use as a response for that intent
• If you have multiple articles, ensure that you link all of them and decide which entity matching condition sets should apply to each article to be matched

• Once you have completed the above step, make sure to:

  • Retrain your intent model

  • Optional - add test phrases for each individual intent, preferably questions that your customers asked as shown in your recognition analysis dashboard

Congratulations! You can now enjoy all of the benefits of release 4.8.