The NLP Insights helps you gain in-depth insights into the analytics data and assess your virtual assistant’s performance in identifying and executing tasks. You can improve your VA’s performance based on the insights. The Analyze > NLP Insights page shows the specific information in the following sections:

• Intent Found: Number of identified intents
• Intent Not Found: Number of unidentified intents
• Unhandled Utterances: Number of unhandled utterances
• Failed Task: Number of unsuccessful tasks
• Performance: Monitor all the scripts and API services and capture the number of failed services or scripts during the interactions
• Pinned: Pinned NLP Insight records. Specific records are pinned to highlight them for easy access and viewing.
• Debug Log: Custom Debug logs include user conversations from across all channels, for analyzing your VA.

Intent Found

An intent refers to the goal the customer has in mind when typing in a question or comment. The  phrases used to express the intent are called user utterances. The Intent Found tab includes all the user utterances identified by the platform. 

Note: You need to check for false positives in scenarios where the utterance is wrongly identified for an intent.

See the following table and Features section to know more.

The following is an example of Intent Found:

User: I want to know my order confirmation

VA: Would you like to switch to Track Order

User: Yes

VA: Lets log you in.
      How would you like to go ahead – Log in or Guest?

In the above conversation, the user utterance of ‘knowing the order confirmation status’ is recognized by the VA and successfully mapped to the Track Order intent.

Description of the Intent Found Fields

The following table lists the fields on the Intent Fount tab with descriptions:

Fields	Description
Utterances	The utterances that are identified by the VA. The details on the tab are grouped by utterances based on the similarity, by default. To turn off grouping by utterance, click the Utterances header and disable the Group by Utterances option.
Intent	The intent/task that is identified. The details on the tab can be grouped by intent. By default, the  Group by Intent option is turned off. To turn it on, click the Intent header and enable the Group by Intent option. If the intent has been identified through a Dialog Task, this column displays the task’s name. If the intent is answered using the Answer from Document feature, this column mentions Answer from Documents.
Traits	All the traits that are identified for the listed utterances. The details on the tab can be grouped by traits. By default, the  Group by Traits option is turned off. To turn it on, click the Traits header and enable the Group by Traits option. This information is available for the data generated after June 1, 2021.
UserID	The UserID of the end user related to the conversation. You can view the metrics based on either Kore User id or Channel User Id. Channel-specific ids are shown only for the users who have interacted with the VA during the selected period.
Language	The language in which the conversation occurred. If it is a multi-lingual VA, you can select specific languages to filter the conversation that occurred in those languages. The page shows the conversations that occurred in all enabled languages by default.
Date & Time	The date and time of the chat. You can sort the data by either Newest to Oldest or Oldest to Newest.

Intent Not Found

Intent Not Found includes all the user utterances that the platform is not able to identify with a dialog task or FAQ either due to invalid training, less training data or the intent unavailability in the virtual assistant.

See  the following table and Features section to know more.

Example of Intent Not Found: 

User: I want to know my account statement

VA: I’ m sorry, I did not recognize the value you have entered. Please select a value from the list.

In the above conversation, the VA does not recognize the user utterance of ‘knowing the account statement’. It could be due to invalid training, less training, or unavailability of intents in the virtual assistant.

Description of Intent Not Found Fields

The following table lists the fields on the Intent Not Found tab with descriptions:

Fields	Description
Utterances	The utterances that are not identified by the VA. The details on the tab are grouped by utterances based on the similarity, by default. To turn off grouping by utterance, click the Utterances header and disable the Group by Utterances option.
Traits	All the traits that are identified for the listed utterances. The details on the tab can be grouped by traits. By default, the  Group by Traits option is turned off. To turn it on, click the Traits header and enable the Group by Traits option. This information is available for data  generated after June 1, 2021.
UserID	The UserID of the end user related to the conversation. You can view the metrics based on either Kore User id or Channel User Id. Channel-specific ids are shown only for the users who have interacted with the VA during the selected period.
Language	The language in which the conversation occurred. If it is a multi-lingual VA, you can select specific languages to filter the conversation that occurred in those languages. The page shows the conversations that occurred in all enabled languages by default.
Date & Time	The date and time of the chat. You can sort the data by either Newest to Oldest or Oldest to Newest.

Unhandled Utterances

The Unhandled Utterances help analyze  the unidentified inputs received from the users during a task execution at an entity node, message node, or confirmation node. These insights allow you to identify the need for additional training or new intents and enhance the existing NLU model of the VAs.

In an unhandled utterance, the following new fields are available:

• Prompt Type – A prompt type could be either an Entity node, Message node, or a Confirmation node.
• Node Name – Name of the node in which the utterance is not handled
• Task Name – Name of the task in which the utterances are unidentified on an entity, message, or confirmation nodes.

• Group by functionality is available for Utterances, Traits, Prompt Type, Task Name, and Node Name

Note: The unhandled utterances are available for all conversations with product version 9.3 or higher.

See the following table and Features section to know more..

Unhandled Utterance Examples

The following examples show the conversations between the VA and user, with unhandled utterances captured at Entity, Message, and Confirmation nodes.

At an Entity Node:

For example, when the user provides an invalid input at entity or confirmation nodes as follows:

User: I want my account statement

VA: Please enter your Customer Id

User: Where do I find it?

VA: Sorry, that is an incorrect input. Please enter your Customer Id

In the above conversation, if the VA doesn’t recognize “where to find customer id” as an intent or entity, then this utterance is  categorized under Unhandled Utterances.

At a Message Node:

For example, when the user provides an invalid input at a message node as follows:

User: I want to book a flight for today

VA: Enter the flight number
            User: 12434
            VA: Enter number of seats required
            User: 3
            VA: Your flight is booked. Would you like to:
                    1. Book a Hotel
                    2. Book a sightseeing tour
            User: I want to Cancel the Flight
            VA: I’m sorry, I don’t understand. Please enter again.

In the above conversation, if the VA doesn’t recognize the intent name “cancel flight” as an input at the message node. The intent identification fails and the utterance is categorized under Unhandled Utterances.

At Confirmation Node:

For example, when the user provides an invalid input at a confirmation node as follows:

VA: How may I help you

User: I want to book a flight

VA: Enter the number of seats

User: 2

VA: Please confirm if you want two seats

User: I want to hire a cab

VA: I cannot understand it, can you rephrase it

In the above conversation, when a user enters ‘I want to hire a cab’ at the confirmation node, it is not recognized and logged under Unhandled Utterances.

Description of Unhandled Utterances Fields

The following table lists the fields on the Unhandled Utterances tab with descriptions:

Fields	Description
Utterances	The unhandled utterances for which the inputs received are unidentified. The details in the tab are grouped by utterances based on the similarity, by default. To turn off grouping by utterance, click the Utterances header and disable the Group by Utterances option.
Traits	All the traits that are identified for the listed utterances. The details in the tab can be grouped by traits. By default, the  Group by Traits option is turned off. To turn it on, click the Traits header and enable the Group by Traits option. This information is available for analytics generated after June 1, 2021.
Prompt Type	A prompt type could be either an Entity node, Message node, or a Confirmation node. To turn on grouping by Prompt Type, click the Prompt Type header and enable the Group by Prompt Type option.
Task Name	The task that is identified for the user utterance. To turn on grouping by task name, click the Task Name header and enable the Group by Task option.
Node Name	The name of the service or script or WebHook within the task that got executed in response to the user utterance. To turn on grouping by node names to which these scripts or services belong, click the Node Name header and turn on the Group by NodeName option.
UserID	The UserID of the end user related to the conversation. You can view the metrics based on either Kore User id or Channel User Id. Channel-specific ids are shown only for the users who have interacted with the VA during the selected period.
Language	The language in which the conversation occurred. If it is a multi-lingual VA, you can select specific languages to filter the conversation that occurred in those languages. The page shows the conversations that occurred in all enabled languages by default.
Date & Time	The date and time of the chat. You can sort the data by either Newest to Oldest or Oldest to Newest.

Failed Task

In a scenario where all the user utterances are successfully mapped to an intent, but the task cannot be completed for some reason, then such utterances are listed under this tab. You can group them based on task and failure types to analyze and solve issues with the VA.

See the following table and Features section to know more.

Failed Task – Type of Issues

Different types of issues that occur during a Failed Task are listed as follows:

• Task aborted by user
• Alternate task initiated
• Chat Interface refreshed
• Human agent transfer
• Authorization attempt failure – Max attempts reached
• Incorrect entity failure – Max attempts reached
• Script failure
• Service failure
• Inactivity or External Events (from ver8.0) – when the conversation session and as a result, the in-progress task is closed due to inactivity or external events.

Description of Failed Task Fields

The following table lists the fields on the Failed Task tab with descriptions:

Fields	Description
Utterances	The utterances that are successfully mapped to an intent, but still the task failed due to some issue. See Failed Task – Type of Issues for more details. The details in the tab are grouped by utterances based on the similarity, by default. To turn off grouping by utterance, click the Utterances header and turn off the Group by Utterances option.
Task Name	The task that is identified for the user utterance. To turn on grouping by task name, click the Task Name header and enable the Group by Task option.
Failure Point	Nodes or points in the task execution journey where the failure occurred resulting in the task cancellation or user drop. Click an entry to view the complete conversation for that session with markers to identify the intent detection utterance and the failure/drop-out point. Depending on the task type, clicking Failure Point shows more details.
Type of Issue	Shows the reason for failure in case of Task Failure records. To know the usual type of issues, see Failed Task – Type of Issues.
UserID	The UserID of the end user related to the conversation. You can view the metrics based on either Kore User id or Channel User Id. Channel-specific ids are shown only for the users who have interacted with the VA during the selected period.
Language	The language in which the conversation occurred. If it is a multi-lingual VA, you can select specific languages to filter the conversation that occurred in those languages. The page shows the conversations that occurred in all enabled languages by default.
Date & Time	The date and time of the chat. You can sort the data by either Newest to Oldest or Oldest to Newest.

Performance

Developers can monitor all the scripts and API services across the VA’s tasks from a single window. The performance tab displays information related to the backend performance of the VA. The platform stores the following meta-information:

• Node name, type, and task name
• Total number of runs
• Success %
• The total number of calls with 200 responses and the total number of calls with a non-200 response. You can view the  actual response code from the details page that  opens when you click the service row
• Average Response times
• Appropriate alerts if a script or a service is failing consecutively

Description of Performance Fields

The following table lists the fields on the Performance tab with descriptions:

Fields	Description
Node Name	The name of the service or script or Webhook within the task that got executed in response to the user utterance. To turn on grouping by components to which these scripts or services belong, click the Node Name header and turn on the Group by NodeName option.
Type	Shows whether it is a script or service or Webhook. Webhook details are included from ver 7.0.
Task Name	The task that is identified for the user utterance. To turn on grouping by task name, click the Task Name header and enable the Group by Task option.
Total Runs	The total number of times within the date period that the script or service is run for any user utterances.
Success%	The percentage of the service or script runs that got executed successfully.
2XX Responses	The percentage of the service or script runs that returned a 2xx response.
Non 2XX Responses	The percentage of the service or script runs that returned a non-2xx response.
Avg Response Time	The average response time of the script or service in the total number of runs. This can be sorted from High to Low or Low to High under the Performance tab.

Pinned

Any records from Identified and Unidentified Intents, Unhandled Utterances, or Failed Tasks tabs that are pinned are displayed in the Pinned tab. 

Description of Pinned Fields

The following table lists the fields on the Pinned tab with descriptions:

Fields	Description
Utterances	The pinned utterances are displayed here. The details in the tab are grouped by utterances based on the similarity, by default. To turn off grouping by utterance, click the Utterances header and disable the Group by Utterances option.
Intent	The intent associated with the pinned utterance. The details in the tab can be grouped by Intents. By default, the  Group by Intent option is turned off. To turn it on, click the Intent header and enable the Group by Intent option.
Type of Issue	Shows the reason for failure in case of Task Failure records. To know the usual type of issues, see Failed Task – Type of Issues.
UserID	The UserID of the end user related to the conversation. You can view the metrics based on either Kore User id or Channel User Id. Channel-specific ids are shown only for the users who have interacted with the VA during the selected period.
Language	The language in which the conversation occurred. If it is a multi-lingual VA, you can select specific languages to filter the conversation that occurred in those languages. The page shows the conversations that occurred in all enabled languages by default.
Date & Time	The date and time of the chat. You can sort the data by either Newest to Oldest or Oldest to Newest.

Debug Log

Any custom debug statements that you entered in the Script node using the script koreDebugger.log("<debug statement>")are displayed on this tab. Debug statements should be in a string format. See the following table to know more.

The logs include the user conversation from across all channels. You can use them for bot analysis especially in case of failures during user interaction.

The details include:

• The actual statement that you have defined at the time of Bot definition.
• Date and time of logging
• Channel
• User ID (along with channel-specific id)
• Language of interaction
• Task name if available

You can also view the details of the chat history associated with the session. To view more details, follow the below steps:

1. Click a logged record.
2. On the corresponding window, you can find the Details and Chat History tabs.
3. Under the Details tab, you can find the task name, channel, language, and flow.
4. Click the Chat History tab. You can find the chat transcript where the log is recorded.
   • If the debug log is generated from a VA message, you are navigated to that specific message in the chat transcript.
   • If the debug log is not part of the VA message, you are navigated to the latest message added before the debug statement.

For universal VAs, the debug statements from the universal and linked assistants are included in the logs. The debug logs also include the error messages related to BotKit, for example, when the platform could not reach the BotKit or when the BotKit did not acknowledge the message sent by the platform. The message includes details like the <endpoint>, <error code>, and <response time>.

Description of Debug Log Fields

The following table lists the fields on the Debug Log tab with descriptions:

Fields	Description
Log	Description of the debug log. For example, getIndex is not defined.
Task Name	The task that is identified for the user utterance. To turn on grouping by task name, click the Task Name header and enable the Group by Task option.
Debug Point	The point or a node in the conversation where the error is identified. For example, buildDataForCarousel
Channel	Specific channel where the conversation occurred.
Language	The language in which the conversation occurred. If it is a multi-lingual VA, you can select specific languages to filter the conversation that occurred in those languages. The page shows the conversations that occurred in all enabled languages by default.
UserID	The UserID of the end user related to the conversation. You can view the metrics based on either Kore User id or Channel User Id. Channel-specific ids are shown only for the users who have interacted with the VA during the selected period.
Date & Time	The date and time of the chat. You can sort the data by either Newest to Oldest or Oldest to Newest.

Storage Limitations

The platform imposes restrictions on the number of log statements retained per VA. The limit is a combination of volume and period:

• Only the latest 700 statements per VA are stored.
• Statements older than 7 days are removed.

NLP Insights Analysis

The following sections describe more about the options available on NLP Insights page and the analysis of the records captured here.

Features

The following list details the features available in NLP Insights for Intent Found, Intent Not Found, Unhandled Utterances, and Failed Task.

• You can filter the information based on various criteria such as User Utterances, Intent, user id (Kore user id or channel-specific unique id), date-period, the channel of use, language, etc. You can also filter records based on multiple custom tags. See Filter Criteria to know more.
• Complete meta information is stored for later analysis, including the original user utterance, the channel of communication, entities extracted (if any), custom tags applied, detailed NLP analysis with scores returned from each engine, and the ranking and resolver scores.
• Ability to view the chat transcript to the point of the user utterance. This also gives the option to view the user profile and the details for that user’s conversation sessions.
• You have an option to train the utterance.  The utterance will be marked once trained.
• Any important record you want to mark, track later, or both can be pinned. They l appear on  the Pinned tab.
• Sorting feature is available for Date and Time (Oldest to Newest, Newest to Oldest). You can export the insights data as a CSV file.

Note: The NLP Insights page shows the conversations from the last 24 hours by default. You can filter the insights for a selected period – use the Date drop-down to select 24 hours, last 7 days, or a custom period.

Fields Matrix

The following matrix shows the availability of fields on each tab of NLP Insights:

Fields	Intent Found	Intent Not Found	Unhandled Utterances	Failed Tasks	Performance	Pinned	Debug Log
Utterances	✔	✔	✔	✔	X	✔	X
Intent	✔	X	X	X	X	✔	X
Traits	✔	✔	✔	X	X	X	X
UserID	✔	✔	✔	✔	X	✔	✔
Language	✔	✔	✔	✔	X	✔	✔
Date & Time	✔	✔	✔	✔	X	✔	✔
Prompt Type	X	X	✔	X	X	X	X
Task Name	X	X	✔	✔	✔	X	✔
Node Name	X	X	✔	X	✔	X	X
Failure Point	X	X	X	✔	X	X	X
Type of Issue	X	X	X	✔	X	✔	X
Type	X	X	X	X	✔	X	X
Total Runs	X	X	X	X	✔	X	X
Success%	X	X	X	X	✔	X	X
2XX Responses	X	X	X	X	✔	X	X
Non 2XX Responses	X	X	X	X	✔	X	X
Avg Response Time	X	X	X	X	✔	X	X
Log	X	X	X	X	X	X	✔
Debug Point	X	X	X	X	X	X	✔
Channel	X	X	X	X	X	X	✔

Filter Criteria

You can filter the information on the Insights page using various filters. You can save the entered filter criteria and set it as the default filter using Save as Default Filter.

The filter criteria differ slightly between different tabs. The relevant filters are applied when you switch between the tabs on the Insights Page. See Dashboard Filter Criteria to know more details.

Detailed View

For all the user utterances listed on  the various tabs such as Intent Found, Intent Not Found, Unhandled Utterances, etc., you can open more details of the user session by clicking the respective record. The record shows the information on the following sub-tabs: Details, NLP Analysis, and Chat History.

Details

The Details tab shows the basic details of the session along with a JSON file that includes the NLP analysis for the conversation.

If the intent has been answered from a document, this section provides the following details:

• Information on the Intent not being identified by the ML, FM, and KG engines. Hence, the user utterance is answered directly from the document;
• The answer presented to the user;
• The document from which the answer was provided;
• A Similarity Score for how similar the user query is to the document content;
• An option to add the query to the Knowledge Graph as an FAQ.

NLP Analysis

This tab provides a visual representation of the NLP Analysis, including intent scoring and selection.  See Testing and Training a Virtual Assistant and Ranking and Resolver for more information.

Chat History

On the Chat History tab, you can access the exact message or conversation for which the record is logged. It shows the entire chat history of the user session.

Chat History provides visibility into the user information by capturing the following details:

• User Profile: Provides a 360-degree view of the user and their usage metrics.
• User Conversation Sessions: Lists all the sessions of the user in the given period with the selected utterance section expanded.
• Go to Selected Utterance: When you click this icon, the selected utterance is highlighted in orange (see the preceding screenshot).

The following user information details are displayed on the Chat History tab:

Functionality	Attribute	Description
User Profile	Kore User ID	User id assigned by the platform
	Channel Data	Data received from the channel, that is the information available in the User Context.
	User Meta Tags	The total number of meta tags associated with the user and key-value pairs for the most recent ones.
	Latest Interaction	Last time the user interacted with the VA.
	Total Conversation Sessions	The total number of interactive and non-interactive sessions registered by the user from the beginning of time.
	Total Conversation Sessions in the Last 30 Days	The total number of interactive and non-interactive sessions registered by the user in the last 30 days.
	*The next few attributes are not displayed if there is no interaction by the user in the last 30 days
	Last 30 Days’ Intent Detection Rate	(Total identified intents / (Total identified intents + unidentified utterances)) * 100 for the utterances over the last 30 days
	Intents Requested	Total identified intents + unidentified utterances
	Intents Identified	Total intents identified
	Last 30 Days Goal Completion Rate	(Tasks success tasks / (Total success tasks + total failed tasks) ) * 100 for the tasks over the last 30 days
	Tasks Initiated	Total success tasks + total failed tasks
	Tasks Completed	Tasks successfully completed
	Recent Conversation Flows	Top 10 popular conversation flows executed by the user in the last 30 days. Popular flows are determined by the number of instances, for which the conversation flow is executed.
User Conversation Sessions	Session Attributes
	Session Start	Session start date and time.
	Session End	Session end date and time.
	Channel	Channel in which the session is initiated.
	Agent Transfer Tag	The session where the user is transferred to an agent. Sessions should be considered even if the user returns to the VA.
	Drop Off Tag	The session where the user dropped off.
	Total Success Tasks	Count of tasks successfully completed in the session.
	Total Failed Tasks	Count of tasks failed in the session.
	Intents Identified	Count of intents successfully identified in the session.
	Intents Unidentified	Count of intents unidentified in the session and list of unidentified intents.
	Conversation Path	The series of tasks initiated by the user in the session.
	Session Meta Tags	Count of the session meta tags used with the details of the most recent custom meta tags displayed.
	Conversation Transcript
	Message Meta Tags	The chat transcript is annotated with message tags for messages with meta-tags associated with them.
	Agent Transfer	Indicates the point of agent transfer at the last message before transfer.
	Drop Off	Indicate the point of drop off at the last message before dropping off.

Advanced Performance Details

Clicking a service or script or WebHook name opens the advanced details dialog for the service, which lists each instance of its run along with separate tabs for successful and failed runs. Analyzing the average response time of different runs gives you insights into any aberrations in the service or script execution. Click any row to open the JSON response associated with the service or script run.

Train the Virtual Assistant

You can train the specific intents and utterances from the Intent Found, Intent Not Found, and Unhandled Utterances tabs. To do so, hover over a row in any of these tabs, and click the Train icon. It opens the Test & Train page, where you can train the Virtual Assistant. For more information, see  Testing and Training a Virtual Assistant.

Data Export

You can export the data present on the NLP Insights page to a CSV file, by clicking the Export icon on the top right corner of the page.

Once you click the icon, the export process starts, and you can see the progress in the Status Tracker dock. The export file is downloaded to your local Downloads folder. The downloaded file has the information specific to the selected tab and the detailed analysis based on the selected filters.

These records also include the Meta Tag information.