CLOVA Chatbot overview

Prev Next

Available in Classic and VPC

CLOVA Chatbot is an interactive AI service that identifies the intention of the user's questions and provides appropriate answers. You can use a chatbot builder provided by NAVER Cloud Platform to easily build a chatbot service that learns the conversation data entered for it to process and responds with appropriate answers to users' questions. CLOVA continues to improve the chatbot's features and research the optimal conversation model in order to produce the chatbot engine that understands the user's intentions behind their questions and provides accurate answers for them.

Applied technology and model
CLOVA is creating an excellent chatbot by analyzing Korean sentences perfectly through learning input using long-accumulated NAVER's data, know-how, the latest deep learning technologies, and a high-performance GPU environment. CLOVA Chatbot’s engine has deep learning technology that reflects the language model features that are applied. It means that it is able to learn similar meanings in context, sentences, and words and to accurately recognize words entered wrongly, such as typos.
NAVER 's natural language processing NLU engine owns the best morphological analysis technology, which performs analyses that consider the linguistic nature of Korean with well-developed postpositions and endings. It provides features to analyze object names and statements for analyzing natural language intention, and to understand and analyze multiple natural languages so that their meanings can be extracted.

Process of natural language processing
The natural language processing engine goes through the process of feature extraction, model score calculation, and model ensemble to come up with the optimal response. When an original sentence, i.e., query, comes in, it extracts and categorizes features in the query, such as word classes, entities, endings, and morphemes, and sends each feature to various models. Each model processes different feature values. One may focus on analyzing word classes to find answers, and another may look for answers, focusing on the entities. A variety of such models is designed to individually calculate scores and then calculate the ensemble score among them to provide the most accurate answer.

CLOVA Chatbot features

  • High correct answer rate for Korean: CLOVA Chatbot with multiple models applied has a higher correct answer rate for the Korean language than other companies. Having tested multiple chatbots from different competing companies with the same test data revealed that CLOVA Chatbot scored higher even when the number of intents increased.
  • Multi-language support: CLOVA Chatbot supports various languages based on an algorithm model optimized for each language's characteristics. Currently a total of 6 languages is supported: Korean, Japanese, English, Chinese, Thai, and Indonesian.
  • Response composite provided: CLOVA Chatbot provides a response composite to give you more flexibility in creating chatbot responses. It doesn't stop at providing just text answers. Various response composites can be set, such as text, button, multi-button, image, and carousel. You can combine multiple composites. They are converted to fit each messenger's specifications if connected to a messenger.
  • Various analysis indicators provided: CLOVA Chatbot provides various analytic indicators to continuously improve the conversation model. The utterance types reviewed in scenario modeling and the way users actually use it in the real service may differ, and the conversations implemented in the chatbot may not represent all users' utterance patterns. Therefore, the statistics data or data for analyzing utterance records of incoming users is provided.

Information on User Guide of CLOVA Chatbot

The user guide of CLOVA Chatbot consists of the following topics that will help you to effectively use the service. The content that users can view in each topic is as follows:

CLOVA Chatbot related resources