Teach Chatbot

From English wiki app24.online
Jump to navigation Jump to search


A newly created chatbot already can greet your customers and respond in case it fails to understand messages. To make responses to customers’ questions as accurate as possible, teach the chatbot by adding various intents and correcting existing ones, and by training the chatbot.

Create Intents

NOTE. Adding one intent is enough to test the chatbot. After the chatbot is operational, add the rest of the intents.

Each intent deserves an individual discussion. As an example, we will teach the chatbot to reply to customers’ questions about delivery.

To add an intent, do the following:

  1. Go to the Dialogflow website. In the menu on the left, check that you configure the right chatbot.
    BotIntents.png
  2. In the Intents menu item, click button +.
  3. Specify the intent name.
    FirstIntens.png
  4. Click Add training phrases.
    NOTE. The Add training phrases button is available only when creating the first intent. If you create the second and subsequent intents, specify a value only in the Training phrases field.
  5. In the field, add training phrases, for example, related to delivery. To add a phrase, press ENTER.
    TrainingPhrases.png
  6. Click Add response.
    FirstIntens73.png
  7. In the Text Response table, add one or several responses. The chatbot will send them to customers asking about delivery. If the table contains several phrases, the chatbot will send one of them in a random manner.
    Responses.png
  8. Save your intent. Click Save.

After you’ve created the first intent, we recommend that you connect the chatbot to the Bitrix24 portal, configure the Open Channel, and test operation of the chatbot.

If the chatbot successfully passes the operation test, add required intents and modify the default ones. For example, teach the chatbot to inform a client that you do not take calls using WhatsApp.

Start training the chatbot after it “communicates” with customers several times.

Modify Default Fallback Intent and Default Welcome Intent

A default intent (Default Fallback Intent and Default Welcome Intent) is the one that is added automatically when creating a chatbot.

change default intents

Default Welcome Intent allows the chatbot to respond to greetings. For example, if a customer sends “Hello” to the chatbot, it will respond with one of the standard greetings such as “Hello”, “Good Afternoon”, etc.

Default Fallback Intent contains phrases that the chatbot will send to a customer if it fails to understand his or her messages.

Open the intents and check the messages that the chatbot will send to customers. We recommend that you modify the responses the way it fits your company. For example, if most of your customers are young people, delete too formal responses.

IMPORTANT. Do not delete Default Fallback Intent or Default Welcome Intent to ensure proper operation of the chatbot.

To disable the default intents, do the following:

  1. Go to the Bitrix24 portal, and then in the menu on the left, click BotApp24 for Open Channel.
  2. On the control panel, click button The settings of the bot.
  3. In the dialog box that opens, in the Messages: Default Fallback Intent, Default Welcome Intent drop-down list, select Switched off:
    Switched off default Fallback Intent

Train Chatbot

Imagine that you created a chatbot, connected it to the Bitrix24 portal, and it began to respond to customers’ messages. Sometimes the chatbot sends correct answers, but sometimes they are off the mark. The chatbot must be trained so that it would make as few mistakes as possible and send relevant answers to customers. Thanks to the training, the chatbot will be able to answer correctly even if customers send messages with mistakes, odd wording, incorrect word endings, etc.

To train the chatbot, do the following:

  1. Go to the Dialogflow website. In the menu on the left, check that you configure the right chatbot.
  2. Click the Training menu item:
    bot training
  3. In the window that opens, you can view conversations between customers and the chatbot. All conversations are grouped. The conversation name is the first phrase that a client writes to the chatbot. To manage a conversation, click its title.
    Trainings dialogs
  4. In the window that opens, the data is displayed in the mapping format, where User says is a user’s phrase, and Intent is the one that the chatbot has chosen.
  5. If the chatbot has chosen a correct intent, select the tick box on the right.
  6. When the chatbot receives a message and fails to understand, it responds with one of the phrases listed in the Default Fallback Intent. In some cases, the chatbot must declare that it cannot understand a message. Select the tick box for such cases. If the chatbot makes chooses an incorrect response, click the Default Fallback Intent name, and then select a correct intent.
  7. If none of the existing intents is appropriate, create a new intent by clicking button Create new.
  8. Read all parts of the conversation. If you find that the chatbot selects incorrect phrases in some cases, correct them.
  9. Save your changes. Click button Approve.
  10. Manage the rest of the conversations the same way.

If you created new intents while training the chatbot, in the Intents menu item, find and open the new intents. In the Training phrases field, you can view a phrase for which the intent was created during the training. Add other same-type phrases, and then in the Text Responses table, specify responses for them. Click Save.