CLOVA Chatbot scenario
Available in Classic and VPC
NAVER Cloud Platform's CLOVA Chatbot provides you with the environment where you can quickly and easily create chatbots. You can use the CLOVA Chatbot service to create chatbots and link them with various channels. You go through the following major steps to create a chatbot: Designing a chatbot, requesting subscription, creating domain, creating conversations, building, testing, and linking with messengers and external services. Each step requires detailed settings. Please refer to the detailed instructions at each step.
Request subscription to service
To use the CLOVA Chatbot service, you must complete the request for subscription in the NAVER Cloud Platform console. The Terms and Conditions of this service explicate the terms regarding the retention and use of data generated in the CLOVA Chatbot service, consignment of personal information, and obligations of the company and customers. Please make sure to check the details of the Terms and Conditions before planning your service.
This is the step where you plan the chatbot's purpose, service response range, main scenario, etc. Design the chatbot's response range, conversation structure, main scenarios and failure scenarios, APIs, etc.
Domain is a base unit of a chatbot service. The first step to create a chatbot is to create a domain. For example, if you create a chatbot that is responsible for "NAVER Cloud Platform's customer support," the unit containing all the conversations of the chatbot is a domain. From creating conversations to linkage of channels and statistical information, they are operated and managed based on a single domain. CLOVA Chatbot supports multiple languages, and the characteristics of each language are reflected in natural language processing and learning.
Chatbot builder is a web console for inputting and testing chatbot conversation datasets. It provides various features required for developing chatbots. In the chatbot builder, a conversation is the most basic data unit. A conversation consists of a pair of question and answer. A conversation model learns data called conversations, which consist of pairs of question and answer. Then, it responds by finding the most appropriate answer when a user's question is entered.
Create conversations and entities, collect expected utterances from the users, create common messages, connect conversations, and develop APIs.
Natural Language Processing (NLP) which is a basic method of natural language analysis, analyzes morphemes of sentences entered in the question and answer data. Then, Natural Language Understanding (NLU) determines what the sentence means, and which answer is most closely associated with it. In addition, the model learning proceeds with general entity mapping using NAVER's data dictionaries. If only the entities required to be learned are tagged in a specific domain, then the chatbot engine learns from those tagged entities.
CLOVA Chatbot learns from conversation datasets. Therefore, it is recommended that you add three or more regular question data for conversations whose conversation status is ON, and the question data must be added in a language that matches the natural language set for the domain. If you proceed with learning with only one conversation registered, then it may fail because the number of datasets is too low. Make sure to enter three or more conversation datasets for testing. The learning is usually completed in 5 to 10 minutes for 100 conversation datasets or less. It may take a few minutes to several hours, depending on the amount of data.
NAVER Cloud Platform's chatbot builder uses GPU for fast learning. Learning models in various levels are provided rather than learning with just one model. For example, you can test to see that it correctly returns answers from the model you currently requested to build from the Test immediately after build menu.
Sufficient testing needs to be performed in order to increase the user satisfaction for a chatbot. Once model learning is completed, a validation step is required to test whether it operates as the chatbot producer intended, and whether there are any items that need to be improved. For this, NAVER Cloud Platform's chatbot builder provides an advanced test tool, and you can select a test environment to test the learned conversations yourself.
Once a chatbot is created, tested, and deployed, you can link the chatbot with other messengers or with external solutions such as knowledge database and LINE Pay.
Relearning is a process of understanding what conversations the user had with the chatbot, based on the user's conversation records, and reflecting the user's questions the chatbot couldn't answer. For the description about relearning, refer to Relearn.
For the description about relearning, refer to Relearn.