Action usage

AI action helps user to create a comprehensive conversation without too much effort. MindBehind supports three AI providers: Google DialogFlow, IBM Watson and Microsoft Luis.

After connecting to a provider, the page fills up with details. Here, there are two titles: intent and entity.

Intent is the purpose of the conversation, it represents the customers’ reason for writing to the assistant.

To create an intent, the first step is to give this intent a name and click on the create intent button.

After creating the intent, by clicking on this intent, users view the section where they can feed this intent. Here, under the title “User Says”, we can give alternative versions of the same purpose. It’s a simple rephrasing process. Here, whatever you add to the user says part joins to this intent and can all be linked to somewhere in the flow at once.

An entity, on the other hand, is the information we are trying to extract from the conversation.

As an example, we can name an entity as “dinner”. This is a general name for types of dinner we would like to cover.

Then on the intents, we shall click on the intent that is relevant to dinner. Here, our aim is to understand that this user in the conversation wants to order dinner. At the same time we are getting what they want to order specifically.

Here on the defined parameters part we should select “dinner” as an entity and give the parameter name dinner again.

After adding dinner as an entity for this intent, by double clicking on steak, pizza, and pasta we can connect these words to entity “dinner”.

Now, these words have joined the entity dinner.

To define typos and understand the user even when they do not spell the word correctly, we should click on the entity dinner and add these types of dinner as a label. In the synonyms part, it’s possible to give synonyms and typos.

The prediction threshold shows the matching percentage of what the user says to the defined intents. Here, 1 represents a 100% match and 0.5 shows a 50% match. Configuring this number will give you the opportunity to determine the power of your AI.

Conversation Report is an analysis page where the user can track the conversations and see what the user says and how the chatbot responded.

Here, it’s possible to track what the user said, which intent is matched with how much similarity. Also here, we can add intents using the input came from the customer and make other changes to improve AI.

Behavior

AI action is used to interpret user input in a smart way. It uses one of the AI providers to achieve that. This action can play with two modes:

• Message mode: Where it will process the input and continue the flow depending on the result
• Input mode: Where it will stop the flow and wait for an input, after input from the user it will process the input and continue the flow depending on the result

The AI action will only process text messages, in case of a linked clicked message it will do nothing, and in case of other types, it will fallbacks.

In the case of the text message, the AI processor will behave with the next step in order:

1. Will check a user input match in global keywords, and if the match exists it will continue the flow without waiting to the connection specified in that global keyword.
2. Will predict the user input, and if there are a matched intent and the prediction threshold above the action threshold, it will use its connection and continue the flow without waiting. If the prediction threshold under the action threshold, the action will fallbacks.
3. Will fallback if it couldn't catch any match, in that case, if the fallback count defined it will keep sending the local or global error message until the count is finished then it will continue the flow to the fallback connection.

Defining a delay for this action will delay the prediction process which will conclude delaying the flow to the next module.