The Conversation training tool offers a number of features to help you analyze conversations and improve your virtual assistant. This page describes the main features.

Response statistics

The response statistics in the top bar helps to quickly view the amount of responses and good answers, and to see the ratio of reviewed and not reviewed conversations.

Bot has a response: Kbot has at least one proposal on the user statement.
Bot has a good answer: this chart shows whether the bot can provide useful answers. Good answers are the ones that are marked with the Star icon. The calculation takes into account the number of good responses in regard to reviewed chats only.
Live chats reviewed: the ratio between reviewed conversations and the ones that must be reviewed. A conversation is considered reviewed once you define a good answer with the Star icon.

Modifying user statements

Sometimes the initial user statement might be inaccurate or contain excessive details. Modify such messages to make them more relevant for machine learning purposes. Here are some cases when this feature comes in handy:

A user writes that they want to speak to an agent but does not actually describe the issue. Once you have studied the conversation, you know the actual issue, so you can modify the user statement to make it meaningful.
A user statement has numerous “social chatting” patterns, such as “Hey, how are you doing, Kbot? I’ve got a little question, can you help me?” Make the statement clear by removing words and phrases that are not directly associated with the described issue.
A user provided their personal data, such as an account ID.

To modify a user statement, click Edit user statement.

Recalculating insights

Once you have modified user statements, or added intents, or added articles in the Expert training view, it is required to recalculate a conversation’s insights. To do so, click Recalculate insights.

Adding insights manually

If you know that a particular insight really matches a given user statement, you can add this insight manually. In the Insights tab, it is possible to add intents, short answers, or articles.

Intent

An intent is a user's goal, something that they need at the moment. Once your virtual assistant has detected an intent, it responses with the corresponding reply, which is defined in the intent itself. This reply helps the user to solve a given problem.

To add an intent as an insight:

In the Insights tab, click Add intent.
Find the appropriate intent from the list and click it.
Confirm the selection by clicking Select.

Short answer

Short answers might come in handy when you need to provide users with a brief, yet useful tip straight to the point.

To create and add a new short answer as an insight:

In the Insights tab, click Add short answer.
Click Add. The short answer editor opens.
From the Knowledge base drop-down list, select the base the answer must be added to.
Select the language in the upper right of the pane.
Write the title. The title is displayed in the list of insights, but also the title is the first learning phrase. Kbot uses it to detect the short answer.
Provide the content of the short answer.
Click Save.

In the Short answer view, you can view the associated learning phrases. The first item in the list of learning phrases is the title that you define for a short answer. Others are added as you train the bot.

Article

A knowledge base is either an external or internal system that contains information. Kbot provides users with this information to help them solve their problems: there are articles that describe how to deal with technical issues, account problems, configuration features, and so on.

Add articles from your knowledge bases to help a user quickly find the solution.

To create and add a new short answer as an insight:

In the Insights tab, click Add article.
Click Add. The search pane opens.
Specify whether you want to search by a phrase or by an article ID.
Select the knowledge base where the bot must perform the search. By default all of them are selected.
Find the appropriate article and select it by clicking Select an article.
To confirm the selection, click Select at the bottom of the pane.

Reviewing conversations

Track conversation to better understand what is understood by the bot and how to improve it. See the conversation details in the Conversation tab. You can see both user messages and the names of the buttons that a user clicked during a conversation.

Viewing training history

In the History tab, you can track the history of training to learn who marked or trained replies.

Applying filters

In the Filter pane, narrow your search by specifying the following details:

Define the chat scope, or the conversations source. You can view the chats with Kbot or operators, but if your environment is configured to support external sources (such as conversations from ServiceNow, for example), it is possible to select them as well.
Specify the time period of your interest.
Select the status of the conversations that you want to view.
Select the language of conversations.
Search for conversations by specifying their participants: agents and users.

Features