CLOVA Chatbot FAQs

Available in Classic and VPC

Q. There's an error code, and I don't understand what it means.
Please refer to Error codes for the explanations for error codes.

Q. How do I access CLOVA Chatbot service from the console?
From the NAVER Cloud Platform console, click Services > AI Services > CLOVA Chatbot.

Q. What is the difference between NAVER Cloud's CLOVA Chatbot service engine and others?
NAVER Cloud's CLOVA Chatbot service engine is a natural language-based chatbot engine developed with NAVER's accumulated know-how. The conversation modeling engine embedded in the chatbot service uses natural language processing technology and machine learning-based "self learning" to enhance the conversation model, enabling your chatbot to talk with users more naturally.

Q. I would like to know how actual users are using chatbots in terms of usage patterns and utterance types.
How actual users use your chatbot service may differ from what you expected at the conversation creation stage. This is why it's important to analyze the statistical data and history of incoming conversations. NAVER Cloud Platform provides various indicators that can be used to constantly enhance the conversation models. Please refer to Indicators and analysis for more details.

Q. It's my understanding that Google and Watson uses intent-based chatbots. Does NAVER chatbot use pattern matching that requires certain words to be included in order to produce responses?
In the context of chatbots, intent refers to the user's intention. Many chatbot builders use the intent-based approach. Intents can be categorized (or learned) based on what the users entered, based on which chatbot services produce their responses. CLOVA Chatbot supports intent-based approach in the form of conversation types.
Furthermore, CLOVA Chatbot uses a natural language processing engine to create models and learn in a way that resembles how people learn to speak a language. In other words, it studies various sentences and grammar and remembers important information. It also learns repeatedly by looking at the context before and after each interaction and provides feedback focusing on correct answers. It then extracts the answer that appears most frequently as correct in various models. The above is only a brief explanation, but the machine learning process to understand natural language and produce responses is actually highly complicated.
The pattern matching method where the exact word must appear in the conversation is part of its processing techniques. However, this is only a minute part in our overall process, which is an incredibly advanced learning process to process natural language.

Q. Do you have to be a developer to create a chatbot?
You don't need to be a developer to use the chatbot builder to create a conversation dataset-based chatbot available to be produced in the NAVER Cloud Platform console. You can build a conversation model using the chatbot builder and finish chatbot training and deployment in a matter of minutes to hours. You can use the test feature to easily check that the chatbot functions as intended. If you add new utterances for relearning, then you can immediately service the newly learned model as soon as the build is completed.
Note that you may need a developer when CLOVA Chatbot requires using ActionMethod to access external servers, or when it requires integration with custom channels to provide a service. In such cases, the provided guides and chatbot custom API specs must be reviewed with your developer.

Q. How do I provide a voice chatbot?
You can expand your chatbot service with various APIs provided by NAVER Cloud. Especially, the easy integration with CLOVA API (CSR/CSS) allows you to expand your chatbot to voice chatting.

Q. Can I link my chatbot with commercial messenger services such as LINE and TalkTalk?
Incoming conversation messages from various channels are processed in real time to enable linkage with major messenger platforms including LINE, TalkTalk, Facebook, etc. NAVER Cloud's CLOVA Chatbot service supports linkage with not only messenger services, but also with webpages and mobile apps through custom linkage.

Q. I'd like to test my chatbot in a variety of ways. Is there a convenient way to enter conversations?
Bulk uploading using an Excel template is currently supported. A built-in template is also provided, enabling you to easily create simple conversations using the chatbot. Approximately 150 conversation datasets are provided by default, which would be useful for testing your chatbot.

Q. Is it possible to set up a test environment so that a certain amount of conversation data is added in advance? I would like to test how naturally the chatbot can converse if there is enough data.
A built-in template is provided by default, which includes approximately 150 small talk conversations. This is a sufficient amount to be useful for first starting your chatbot. If you create a domain, and have your chatbot learn from one or two scenarios registered, then the build will fail. It is recommended that you enter at least 10 scenarios before conducting a build learning test.

Q. Can I view user IDs when linking with messenger services?
Commercial messenger services such as LINE and Facebook do not provide user's personal information (such as user ID) when providing service linkage via API. Therefore, the user ID value received from the messenger service means the unique value that guarantees that the user is indeed the same user, but can't be used to identify a specific user. The user ID value received from a messenger service is included in the header when calling an action method URL, so you can check the X-KAA-USERKEY value in the header.
X-KAA-USERKEY: e579cf601ea0e6d53ff6d632ab35c720weoir13
If you have to provide a service where the user's exact information needs to be saved, then the best approach is to develop a feature such as separate login to get the information.

Q. Multilingual support inquiry - If an overseas user asks a question in Chinese, can the question be translated to Korean so that chatbot can answer it? Can the answer also be translated from Korean to Chinese for the user to see?
The chatbot supports Korean, English, Chinese Simplified, Chinese Traditional, Japanese, Thai, and Indonesian. You can select the language for natural language processing when creating a domain. If you enter conversation datasets for the selected language set, then you can create a chatbot with multilingual support. You can also create conversations in Korean, then add further development so that various Papago APIs (for translation) are linked to your application.
By default, CLOVA Chatbot does not support speech recognition and Papago API linkage. However, as CLOVA Chatbot currently provides the custom linkage feature, you can develop your chatbot further to expand to voice chatbot through CSS/CSR linkage, or to a translation chatbot through Papago API linkage.

Q. I linked my chatbot to LINE and invited the LINE English translation bot, but then the LINE English translation bot exited the chat room. Can't I use the translation bot with a chatbot?
You can't use a translation bot when talking to a chatbot. This is because two bots (chatbots) can't exist in a single chat room. When the answer by the first bot is input into the second bot, and the answer of the second bot is input into the first bot, it can lead to an infinite loop.

Q. Regarding "images and multi-links," is there a way to set an answer to be displayed and scrolled horizontally, like a carousel?
CLOVA Chatbot service supports carousel. However, carousel may not be supported by the messenger service to which you linked the chatbot. Please test on the channel (messenger) before linking.

Q. Build failed.
The learning time depends on the amount of data, but it usually takes 5 - 10 minutes if the number of conversation datasets is 100 or less. If you create only one conversation and start learning, then learning may fail because there aren't enough datasets. You should enter at least 10 datasets before testing. You won't be charged for failed builds.

Q. I want the chatbot to greet the user before user says something when the user opens the chat window.
You can use the "Welcome message" feature from the chatbot builder's Common message menu. It can only be set up if the channel (messenger service) you're going to link supports it. Once a welcome message is set up, the chatbot will send the user a designated welcome message when the user begins chatting with the chatbot. However, in the case of building a custom Chatbot, the welcome message feature must be developed separately so that it operates according to the service plan.

Q. I'd like to make my chatbot continue the questions and answers based on the previous answers, which the user selected from multiple choice questions by using forms. For example, "What kind of pizza would you like to order, Option 1 (pepperoni), Option 2 (super supreme)?" If the user chooses Option 1 for this question, the next question is "Select the type of ham for the pepperoni pizza from the following Options. Option 1 (Belgian ham), Option 2 (Jinju ham)" and so on. How do I make the chatbot ask different questions depending on the answer to the previous question?
You can use slot conversations. For more details, refer to the slot conversation section of the Task guide.

Q. I'd like to retain the answers and send several answers to an external API at once. How can I retain the answers?
You can use slot conversations in task to retain answers, so that you can send them later at once using an action method.

Q. How do I create answers when two questions are included in a single sentence?
Write the answers to the two questions. For example, if user asks "how much are the server fees, and how I can use it?" you prepare the answer regarding the server fees and how to use it.

Q. Is it possible to copy conversation datasets from Domain A to Domain B?
You can conveniently manage your conversation datasets by using the Copy domain and Upload/Download Excel features.

Q. How long does it take to build a chatbot?
The build time depends on the amount of data, but it usually takes 5 - 10 minutes to complete if the number of datasets is 100 or less. If you attempt to build a chatbot having registered only one conversation, then the build will fail because the number of datasets is too low. Make sure to enter at least 10 or more conversation datasets before building.

Q. How can I link my chatbot with custom APIs?
To set up custom API linkage, you must be subscribing to API Gateway. For more details, refer to the Channel linkage guide (basic linkage between the chatbot and API Gateway) and CLOVA Chatbot Custom API Spec guide.

Q. How do I create multi-buttons in my chatbot?
NAVER Cloud Platform's chatbot provides two features that can be selected using buttons.

• URL link button: Text appears in a button, and a new window opens to the designated URL address when the user clicks the button. Among answer types, "Multi-link answer" should be used for multiple URL links.
• Multiple choice button: Choices appear as buttons. When the user selects one, the answer corresponding to the selected button is returned. Use “Multiple choice answer” for the multiple choice form.

Q. Can I use punctuations such as question marks and exclamation points?
The chatbot engine would judge sentences without the punctuations, so when the user enters "Hungry?," it will look for the answer as if the input was "Hungry."

Q. If I copy a domain set as Korean and change the domain's language to another language (Japanese, Chinese), will the chatbot still learn?
The chatbot service recognizes and learns based on the language selected when creating the domain. If Korean conversation sets exist in a domain set as Japanese, then the build will fail.

Q. Functional differences between slot and user variable

Q. How do I change the timeout value of action method calls?
The default value of action method calling timeout is 200 ms, and this can't be changed directly by clients. Please submit your request to our customer center.

Q. I registered a welcome message, but it doesn't appear in the test. What do I do to resolve this?
Please check if you performed the conversation model build or applied the changed settings after registering the welcome message. You can also check if you've clicked the [Start with welcome] button if it was a manual test.

Q. Differences in action method versions

Q. After adding conversations in the chatbot builder, can I perform tests in advance before building? I would like to check if the conversations work properly before I build.
You must complete building before starting a conversation test. We are planning a feature to test added conversations in the conversation list to be released in the future.

Q. There is spacing in the entities I registered, but when I get responses from the recognized scenario the spaces are gone. Is it possible to make the registered entities appear the same (to the spaces) in the scenario's response entity field?
The current specifications don't take spacing into account when processing entities, and the spaces are removed when returning entities. The feature to return the analysis values with the spacing maintained is currently not supported.

Q. I'd like to receive the "number of licenses" and "number of months on the license" as entered by the user via a task's entities, and use this information to answer back with a value of fee (5000 KRW x number of licenses x number of months on the license). How can I perform computations within a basic answer?
The computation feature in the entities themselves is currently not supported. You should make user variables inherit the analyzed entity values and then handle the computation within the user variables. Please note that the computations within the chatbot builder can only provide additions and subtractions. Multiplication is not in the scope, so this needs to be developed separately.

Q. Are the fees billed monthly or weekly?
The fees are calculated monthly.

Q. I want the chatbot to not send any answer, including failure messages, if it's not able to find an answer due to the input analysis being in the low accuracy section.
If you are using a custom channel, then you can check the message type, and set the failure message to not answer the user. For other messenger channels, the feature to not send any answers instead of failure messages is not supported.

Q. If I've set multiple choice answers and multi-link answers at the time, can I designate the priority?
Priorities for specific answers are not supported in CLOVA Chatbot.

Q. I want to use a URL in a chatbot's answer. Can I use a localhost address?
Localhost addresses can't be used.

Q. I want to back up a domain.
You can roll back to the previous build version, even after a build. You can also use the Copy domain feature to copy the domain aside to back up a domain.

Q. I used a Python code to call API to custom link a chatbot, but the result text is not appearing correctly. Are there more to be done other than building the conversation and deploying it?
You should complete the service deployment in the Build history menu for it to respond normally to API calls. Please check if you've only deployed the beta.

Q. The regular expression question I entered in a conversation question doesn't seem to work correctly.
Currently, a regular expression question can be as long as 500 characters. Please make sure your question containing the regular expression is no longer than 500 characters.

Q. How can I make my chatbot to repeat the sentence the user entered, and then provide an answer? For example, if a customer says "I have trouble understanding the subway map," I want the chatbot to first say "You would like help with 'I have trouble understanding the subway map,' right?" and then provide an answer.
If you'd like to use a sentence that the user entered in an answer, you can develop an action method to use in the conversation's answers.

Action method example
def main(args):
userInfo = args.get("userInfo", {"query":""})
query = userInfo.get("query", "")
return {"data":[{"variableName":"query","value":query}]}