Intent Discovery (Beta)

The new Intent Discovery module helps you auto-extract popular intents from previous user conversations. It reduces the time and effort to build a virtual assistant and leads to the success of your Conversational AI Journey. This is a beta feature and is available only for the English language and Enterprise users.

You can upload your historical transcripts in CSV format. After the transcripts are uploaded into the bot, the bot uses LLMs to identify the different topics, intents, or conversations between the user and the bot. It intelligently identifies all the intents available. You can review each intent to understand which conversations have resulted in identifying these intents. After the review, you can also see the underline utterances that resulted in identifying an intent. You can either add these intents as new intents for your virtual assistant or pick specific utterances and train them as utterances for your existing dialogs and FAQs. So, it helps both ways – either create new intents or enhance the training you provide to your virtual assistant.

The Intent Discovery journey consists of the following steps:

1. Create a Project: You may have many business use cases to automate. Projects help manage intent discovery for each business use case separately – a project for a specific use case.
2. Upload Transcripts: Upload your use case-specific historical transcripts in CSV format. Conversations are extracted from the uploaded transcripts.
3. Extract Intents: Intents and training data are extracted from the conversations.
4. Review Intents to create new intents or add them as training data: You can add new intents as dialogs or FAQs or train them for existing dialogs or FAQs.

Create a Project

1. Go to Build > Tools > Intent Discovery.
   
2. Click the Create Project button.
3. Enter a name for the project and a brief description of the project. Note that the character limit for Project Name is 256 and for Project Description it is 1000.
   
4. Click Proceed. The new project is created, and the Transcripts tab is displayed where you can upload the historical transcripts.

The Intent Discovery page lists all the projects with conversational insights for each project, such as status of the project, number of conversations, intents added, intents count, and intents trained.

The following different statuses of the project are displayed on the Status column:

• Conversation In Progress
• Conversation Extracted
• No Files Uploaded
• Intent Extraction In progress
• Intent Extracted
• <#number of> Intents Added
• <#number of> Intents Trained

You can edit the project’s name and description. You can also delete the project if required.

Upload Transcripts

Once a project is created, you are taken to the Transcripts tab to upload transcripts in CSV format. You can download the sample CSV file from the tab. Ensure to upload transcripts in the format specified in the sample file.

Format of the CSV file

Field Name	Description
Id	The session ID of the conversation. (For all the conversations in a session, the Id must be the same.)
Sender ID	The ID of the bot or agent.
Other Party ID	The ID of the user talking to the bot or agent.
Direction	The message direction – in (user input) or out (bot/agent response).
Message	The message shown to the user or the bot/agent, depending on the message direction.
Message Type	Format in which the message is sent or received – text/html.
Created on	Date & time on which the message is sent or received.

Limitations

• Number of projects allowed: 2 projects per bot.
• Number of transcripts files allowed: Up to 5 CSV files per project.
• Maximum size of CSV transcript file: 1MB each.

Steps to upload the transcripts:

1. After you create a project, the Transcripts tab opens. (Alternatively, on the Intent Discovery page, you can click the project for which you want to upload transcripts.)



2. On the Transcripts tab, click the Upload Transcript button, and then browse and select the transcripts files (CSV format). The selected files are uploaded, and conversations are extracted from the uploaded transcripts.




3. Upon successful upload, all the transcript files are displayed under the Transcripts tab.

The following details of the uploaded transcript files are captured and displayed:

• File Name – Name of the transcript file
• Upload Time– Time taken to upload the file
• Number of Conversations – Unique conversations between the agent/bot with end users.
• Average Messages – Average number of messages exchanged between agent and end-users per conversation
• Status – Status of the uploaded file.

Any of the following statuses can be displayed for each of the files that are being uploaded:

• Conversation Extraction In progress
• Conversation Extracted
• Invalid File
• Upload Failed
• Intent Extraction In progress
• Intent Extracted
• Intent Extraction Failed

Delete Transcripts

The uploaded transcript file can be deleted if the file is not required anymore, in cases where the upload was incorrect or if the conversation has been extracted and the file is not useful.

View Conversations in the Transcript

Upon successful upload of the transcript, you can view all the conversations and the chat messages exchanged between the end user and the VA/agent.

The platform displays the conversation session summary, which includes the count of the VA messages and user messages. You need to select a Date Filter under Contents to display the details.

If a conversation is not useful and is not contributing to the intent extraction, you can delete the specific conversation. Deleting a conversation also impacts the conversation count and average message count in the columns.

Extract Intents

The platform can extract the conversations from the uploaded files. After validating all the conversations within transcripts, the next step is to extract intents and training data from the conversations without segregating them manually.

Note: You can extract intents only once per project. Make sure to upload all the required transcripts in your project before initiating the extraction. If you need to extract intents from the same file again, you must create a new project. Intent extraction may take a few minutes based on the number and nature of the conversations.

Steps to extract intents from the transcript file:

1. In the Transcripts page, click the button Extract Intents and click Confirm on the pop-up to proceed with the extraction.
   
2. The intents get extracted and the status can be seen across the file names, as shown below.
   
3. Once the extraction is completed, click the Review Intents button.
   
4. You can see the list of extracted intents under the Intents tab.
   

All the extracted intents are listed with the count of utterances that are extracted from the uploaded files.

5. Click any extracted intent to view all the influenced utterances that have led to intent identification.

The platform displays the list of utterances, mapped intents and status of the utterances (Reviewed/Yet to Review).
Under Mapped Intents, the details of all the intents and FAQs used in training the utterances would be displayed. See Train the Utterances to know more.

6. Click the View icon to see the transcripts of the utterances.
   
   

Train the Utterances

Once the utterances are listed, you can select one or more utterances and perform the following actions:

• Create a new Dialog Task or FAQ
• Train them for an Existing Dialog Task or FAQ
• Change the status of the utterances

Create New Dialog Task

In a scenario where the utterances cannot be mapped to the existing intents, you can create a new intent.

Steps to add a new intent (DialogTask or FAQ) for the utterance:

1. Under the Influenced Utterances, select one or more utterances for which you want to add a new Dialog Task.
2. Select the Dialog Task option from the Add Intent drop-down.
   
3. A new dialog task creation window is displayed.
   
4. You can modify the Intent Name extracted from the conversations if required, and add all other details in the Add Dialog Task panel. Also, you can delete the utterances if they are not valid for the intent.
5. Click the button Add to Bot to add the intent for the utterances. Click Yes, Mark as Reviewed if you want to change the status of the trained utterances to Reviewed, or else click No.
   
6. The trained utterances are displayed as shown below. You can click the View More link to see the details of the mapped intents – all the FAQs and the Dialog Tasks to which the utterance is mapped.
   
7. To view the details, click the View More link under the Mapped Intents column from the main page of extracted intents, as shown below.
   

Add New FAQ

You can train the influenced utterances to create new FAQs.

Steps to add FAQs:

1. Select the utterances for which you want to add FAQs, and click FAQ from the Add Intent drop-down.
   You must add the primary question and the bot response to create the FAQ.
2. The Add FAQ panel is displayed, where you can enter all the relevant details.
   
3. You can click Add Extended Response to add an extended response to the main response and also add alternate responses.
   

Train Utterances for the Existing Intents

You can add an existing intent to an utterance or multiple utterances.

Steps to add an intent to utterances:

1. To map an existing Dialog Task or FAQ, click the Train drop-down.
   
2. Click the Dialog Task option to choose the matching intent for the selected utterance(s).
3. Click the button Add Utterances to add the intent for the utterances.
4. Similarly, click the FAQ option from the Train drop-down to add the matching FAQ for the selected utterance(s).
   

Change Status

You can modify the statuses of the utterances extracted from the files. Once the utterances are added or trained for an intent, a message is displayed to change the status. See Train the Utterances to know more. The status of all the utterances is Yet to review by default. You can change the status of an individual utterance from Yet to review to Reviewed and vice versa.

You can also use the Status drop-down provided in the top right corner of the page to change the status of one or more selected utterances.