If Flow Editor is the backbone of your bot, Training is the flesh. Users will not always follow your designed flow. They might choose to disrupt it by asking a random question totally unrelated to the previous messages. A well trained chatbot will handle more edge cases.
So, Training is the feature that helps your bot to increase the ability to understand user’s input even with typos or different wording.
The more set and phrases you add to your bot over the time, it will become smarter over the time and be able to teach itself to understand more. Thus, Training is suitable for things that are text-heavy such as F.A.Q, Diagnose or small talks.
Please note that Training is more advanced than the simpler and more popular keyword mapping.
To train your chatbot, firstly, you have to enable chatbot integration with Natural Language Processing (NLP) services such as Wit.ai and Dialogflow under
Once your chatbot is connected with an NLP service, you can start training your chatbot.
Set is a group of dialogues that have a common content or purpose. We recommend you to create Set per topic/domain to increase accuracy and bot’s learning speed. For example, Order Set or Job Opportunity Set are better than a generic set like FAQ.
Create a "New Set" in Training
New Set, you will be asked to fill out the following:
Name - Set name for Training Set
Short Description of this set
Language - The language is used in this set. If your bot is multi-lingual, you should create many sets for the same group.
Minimum Confidence - The minimum ratio of confidence for analyzed results. The higher the ratio is, the more accurate the bot's answer is. In the case of training bot small talks, you can set the low ratio in Minimum Confidence (around 30%) for maintaining dialogue between bot and user.
Phrase: Fill up as many phrases as possible under the part "User may say"
Set up phrase under the part "User may say" in Training
Response: Write manually the answers in the cell defined "Write a response" or "Use a block".
In the case of training small talk, you should write manually response under the part "Write a response". And the answer that bot respond to the user is random.
Training bot by "Write a response" under the part "Bot will respond"
We also recommend you to set up response by "Use a block" that can take user back to flow as soon as render to block. In the case of using more than one block, bot can respond randomly one of the answers.
Answer phrase by using a block
Tips for building a smart chatbot like a human
TIP 1 - Give the user the directional cues
We recommend you to design your chatbot which ask relevant questions from the user and provide suggestions that are simple ‘tap to answer’ message buttons defined
Quick Replies. If the user does not know what to say, the chatbot must come up with suggested tasks that it can perform for the user.
This helps in utilizing the chatbot’s capabilities as well as exploring where the bot needs training.
Common Welcoming Message
TIP 2 - Personalise the user interaction
An effective chatbot should probably start with asking the user who he is or how he is feeling etc. Basically, it seems to be good to start the conversation and makes the user feel as he is chatting with a true person.
Taking the user’s name frequently, greeting him, etc. are some practices, which make conversations more personalised.
Common Greeting Message
Personalise the user interaction
TIP 3 - Express your conversation with emotions
The emotion behind your conversation should be in accordance with the purpose of the bot. If the user has accomplished a task, give a cheerful response. If the user could not accomplish a task, feel the inconvenience and convey it in the correct manner.
In almost cases, emotions should be based on context of the situation.
Cheerful response with emotion