CLOVA Chatbot quickstart

Available in Classic and VPC

NAVER Cloud Platform's CLOVA Chatbot service provides an environment that enables the quick, easy creation of chatbots. Through the CLOVA Chatbot service, you can create chatbots and link them with a variety of channels. The steps for creating a chatbot are largely divided into chatbot design, service use application, domain creation, conversation creation, building, testing, and integration with messengers and external services. Because each step requires careful setup, refer to the detailed description of each step.

Request subscription to service

To use the CLOVA Chatbot service, you must first complete the service use application on the NAVER Cloud Platform console. These Terms of Service contain details regarding the storage and use of data generated by the CLOVA Chatbot service, the consignment of personal information, the company's obligations, and the customer's obligations. Be sure to review these terms before planning your service.

Chatbot design

The purpose of the chatbot, its service response scope, main scenarios, etc. are designed during this step. It involves working on features such as the chatbot's response scope, dialog structure design, main scenario and failure scenario design, and API design.

Create domain

As a standard unit for chatbot services, the first step in creating a chatbot is to create a domain. For example, if we want to make a "NAVER Cloud Platform customer support" chatbot, the unit that contains the entire conversation of the chatbot is the domain. From the creation of conversations to the integration of channels and statistical information, everything is operated and managed based on a single domain. The CLOVA Chatbot supports multiple languages ​​and reflects the characteristics of the selected language during natural language processing and learning.

Create conversation

Chatbot Builder is a web console for entering and testing chatbot conversation datasets, and provides various features required for chatbot development. The most basic data unit in Chatbot Builder is conversation. A conversation consists of pairs of questions and answers. The conversation model learns data in the form of registered conversation pairs of questions and answers. Then, when a user inputs a question into the chatbot, it finds the most appropriate answer and responds.
This step involves working on conversation creation, entity creation, collecting expected user utterances, writing out common messages, conversation integration, API development, etc.

Build

Build method
Natural Language Processing (NLP), the basis of natural language analysis, analyzes the morphemes of sentences entered in question data and answer data. The next step, Natural Language Understanding (NLU), determines what the sentence means and which answer it is most similar to. Additionally, as part of learning, general entity mapping tasks are performed by mapping with NAVER's data dictionary. If you tag only the entities that need learning in a specific domain, the engine will learn them including the tagged contents.

Build time
CLOVA Chatbot learns based on conversation datasets. Therefore, it is recommended to register at least 3 general question data in a conversation that is in the service ON state. These must be registered in a language that matches the natural language set in the domain. Learning may fail due to the dataset being too small if learning is carried out with only one conversation registered, so at least 3 conversation datasets must be entered before testing. Learning time is 5 to 10 minutes for conversation datasets of less than 100, and can take from several minutes to several hours depending on the amount of data.

Build resources
NAVER Cloud Platform's Chatbot Builder utilizes GPU for rapid learning processing. Rather than learning with just one model, it provides learning models at different levels. For example, as soon as you finish building, you can test whether the model that requested the current build responds normally in the test menu.

Test

To increase user satisfaction with a chatbot, ample testing must be conducted. After model training is complete, a testing and verification step is needed to check whether the chatbot works as intended by the creator and whether there are any items that require further improvement. To this end, NAVER Cloud Platform's Chatbot Builder provides advanced testing tools, and you can directly test the learned conversations by selecting a testing environment.

Linkage

After creating, testing, and releasing your chatbot, you can link it with other messengers or external solutions such as LINE Pay.

Retraining

Retraining is the process of understanding what kind of conversations a user has had with a chatbot based on the user's conversation history and adding the user's questions that the chatbot was unable to answer into the conversation.