Advanced settings

Available in Classic and VPC

Clicking [Advanced Settings] at the top of the conversation information allows you to configure various detailed features to enrich the conversation. You can set the context to configure multi-turn conversations, or change the conversation type to a general conversation or a task. You can also configure negative data that helps train the model through negative settings, or set feedback response messages separately for each conversation.

The following items can be set in the advanced settings popup window.

Context settings

If you have created a welcome message, multiple general conversations, tasks, failure messages, and so on to build a chatbot, you need to link these conversations, and this is called context setting. Context is a link that connects one conversation to another, and is used to understand the context of the linked conversations. To link conversations, enter the same context in the output and input of the conversations you want to connect.

For context, if the output context of conversation A matches the input context of conversation B, then after conversation A is matched, conversation B, whose context matches, is matched with priority over other conversations that have no context. If there is a conversation linked through a multiple-choice response, the conversation linked to the multiple-choice button is matched with priority over context.
The method for setting context is as follows.

Enter the conversation information by referring to Register conversations.
- The advanced settings button is enabled only after conversation information is registered.
Click [Advanced Settings] in the conversation.
Register the context information on the [Context Settings] tab.

Item	Description
Context intensity setting	Setting the intensity of context rules Hard: Avoid deviating from the conversation outside the set context flow. Soft: Allow deviating from the conversation outside the set context flow.
Input context	Enter input context The conversation model identifies the conversation context by comparing the entered Input context with the context remaining in the user's log from previous conversations. If the context remaining in the user's log matches the Input context of the conversation, it is judged to be a continued context.
Output context	Enter output context This is the context left in the user's log after the conversation ends, and the context is deleted once the set count or time is exceeded. Action: Set how the Output context is processed. Add: Add the context to the user's log. Keep: Keep the context without deducting the count that would have been deducted this turn. Delete: Delete the context from the user's log. Count limit: Limits the lifespan of the context kept in the user's log to a set number of times. The set count is deducted each time a conversation turn passes, and once fully deducted, the context is deleted from the user's log. (However, re-prompts in slot conversations are unrelated to the context's lifespan.) Time limit: Limits the lifespan of the context kept in the user's log by time. Once the set time is exceeded, the context is deleted from the user's log.

Note

While the context is maintained, the collected slot information is remembered, so you can retrieve and use slot information collected from previous conversations to build a chatbot service that handles more complex conversations. For a detailed description of how to retain slot information through context, refer to Use tasks.

Context setting example

For example, assume there are the following four conversations.

Conversation A: A reservation task. A conversation that checks the date, number of people, and time, and then confirms whether to proceed with the reservation as the final response.
Conversation B: A conversation in which the chatbot delivers a reservation confirmation message as a response when the user enters "Yes."
Conversation C: A conversation in which the chatbot delivers a reservation cancellation message as a response when the user enters "Yes."
Conversation D: A conversation in which the chatbot delivers the message "Understood" as a response when the user enters "Yes."
After receiving reservation information through the reservation task, the chatbot runs conversation A, which asks whether to proceed with the reservation. If the user answers "Yes," the chatbot looks for a conversation in which "Yes" was entered as the response to a question. If the conversations are not linked by context, the chatbot randomly answers from among the 3 conversations where "Yes" is entered as the question. However, if conversation A and conversation B are linked by a context named "reservation in progress," entering "Yes" as the response to conversation A moves the flow to conversation B.

Conversation type settings

Depending on the chatbot's service quickstart, you can select either a general conversation or a task.

General conversation: A general conversation is suitable when you want to build a simple conversation that identifies intent from the user's utterance and then responds with a registered answer. Most chatbots can be built as a service using only general conversations.
Task: Select a task if you want to build a conversation with a complex flow that collects slots through interactive conversation with the user and uses the collected slots to perform a specific task. For a detailed description of how to configure a task, refer to Use tasks.

Caution

After an answer has been registered, the conversation type cannot be changed. You must change the conversation type before registering an answer.

Conversation group settings

Select the group the conversation belongs to. If no separate conversation group is specified, the "HOME" conversation group is the default.

Task flow settings

On the task flow settings tab, you can set the flow between tasks and general conversations. For a more detailed description of scenarios using task flow, refer to Use tasks.

Leave a task

This option is enabled in task conversations. If the user asks a question unrelated to a slot while filling in the task's slots, this allows the chatbot to briefly leave the task and look for an answer to respond with in a general conversation. However, it does not match tasks other than those with a multiple-choice form or short-answer form registered as the answer.

Interrupt a task

This option is enabled in general conversations. If the user asks a question unrelated to a slot while filling in the task's slots, this allows the chatbot to briefly leave the task and interrupt with the corresponding general conversation.

Return to a task

This option is enabled in general conversations. This sets the chatbot to leave the task, respond with an answer from the general conversation, and then return to the task.

Negative settings

Negative settings are used to train the conversation model on subtle differences between similar conversations. Select a conversation with a similar registered question, or another conversation that the conversation model cannot distinguish, and register it as negative. Through this, the conversation model distinguishes the difference between the registered negative conversation and the current conversation. You can register up to 20 conversations as negative.

Negative setting example

An example of negative settings is as follows.
chatbot-chatbot-2-4_setting-negative_ko

Feedback response settings

On the feedback response settings tab, you can select a feedback response message that surveys customer satisfaction with the conversation. If nothing is set, the feedback message set as the default response message is delivered. For how to add a feedback response message, refer to Register feedback.
chatbot-chatbot-2-4_feedback_ko

Documentation Index