Skip to content

Latest commit

 

History

History
151 lines (95 loc) · 7.93 KB

tutorial-enum-set-entity.md

File metadata and controls

151 lines (95 loc) · 7.93 KB
Raw

When to use ENUM entities and SET ENTITY actions

This tutorial will explain when you should use ENUM (enumeration) entities and SET_ENTITY actions.

Video

Set Entity Tutorial Preview

What is covered

This tutorial will introduce two new features. A new type of entity called ENUM (short for enumeration) and a new type of action called SET_ENTITY, which can refer to one of these enum values, and as the name implies, sets the entity to this value. As you will learn below, these new features are used together and we will explain what they are and how to use them below. Before we get into the details, it is important to understand what problem these features help solve.

enum_entity_type.png action_set_entity_type.png

Problem

There are cases in conversations where the meaning of words depends on the context. Normally the labeled key words are learned and extracted using a language understanding service, but in these cases those systems may not be able to learn using labeled examples.

Imagine you overhear part of a conversation between people nearby and you only hear the word "Yes." You would not know what the "Yes" is agreeing to or confirming since you didn't hear the question asked before it. The question asked before is the context, which gives meaning to the answer. Similarly since "Yes" is such a common response to many different questions it cannot be learned by providing examples as you would do with Custom Trained entities because then it would learn to label every "Yes" as that entity.

Example

Let's further clarify with the following example:

Bot: Do you like Cognitive Research Technologies? User: Yes Bot: Do you like ice cream? User: Yes

In previous tutorials, we looked at Custom Trained entities and your initial thought might be to create an entity named "likesCogServices" and label the first "Yes" as this entity. However, the system would also label the second "Yes". When we attempted to correct the label of the second "Yes" to "likesIceCream", we would then create a conflict of two same inputs "Yes" meaning different things and would be stuck.

It is in these cases that you need to use ENUM entities and SET_ENTITY actions.

When to use ENUMs or SET_ENTITY actions

Use these rules below to know when to use ENUM entities and SET_ENTITY actions:

  • Detection or setting of entity is context-dependent
  • Number of possible values is fixed (Yes and No would be two values)

In other words, use these for any close-ended prompts such as confirmation questions that always result in Yes or No.

Note

We currently have limitation of up to 5 values per enum entity. Each value uses one of the slots in the current 64 limit. See cl-values-and-boundaries

Example: Bot: Is your order correct? User: Yes

When the possible values of the entity are open-ended and not fixed, you would need to use an alternative feature such as expected entity.

Example: Bot: What is your name? User: Matt Bot: What is your favorite color? User: Silver

These prompts are considered open-ended because they could be answered with arbitrary values.

What

ENUM entities

ENUM entities are created just like the other entities. Similar to "programmatic" entities, you cannot label words as these entities. Instead, they must be set through code or SET_ENTITY actions.

enum_entity_creation.png

Set Entity actions

As mentioned above, "Set Entity" actions simply set an entity to a known enum value. You could achieve the same results by creating an API callback action and using the memory manager to set the entity to a value. E.g. memory.Set(entityName, entityValue). Having to write this code and create these actions would become tedious and hard to manage - so Conversation Learner has special actions to facilitate this work and automatically generate these actions when they are used. Having these as independent actions preserves the ability to compose these without being coupled with other actions or code in your bot.

  • Set entity actions can only be created when referring to a value of an enum entity so you must create an enum entity first.
  • Set entity actions are also "non-await" since they have no visible output and need to be followed up by a "wait" action the user can see.
  • Set entity actions are immutable meaning you cannot edit them after creation.

action_set_entity_creation.png

Automatic action generation

If an enum entity exists in your model, Conversation Learner will create placeholder actions for each of the possible values and make them available to select during training. Upon selection, the action would automatically be created for you.

For example, if I create an enum entity with the values "Yes" and "No":

enum_entity_order_confirmation.png

Even without explicitly creating actions for this new enum you would see two new actions available during training:

action_set_entity_sample.png

Create a Bot using these new features

Requirements

This tutorial requires that the general tutorial bot is running

npm run tutorial-general

We will create a bot to simulate fast food ordering. It will have discrete values for sizes of drinks and fries (SMALL/MEDIUM/LARGE), and confirmation questions with YES / NO answers. Both of these entities satisfy the two rules above of being context-dependent answers and fixed values.

Create the model

  1. In the web UI, click on "Import"
  2. Select the tutorial named "Tutorial-Enum-Set-Entity"

This will navigate you to the model management page. Notice the model already contains a few enum entities and set entity actions.

Create the first dialog

  1. On the left navigation panel, click "Train Dialogs", then click the "New Train Dialog" button.

  2. As the user type in, "Hi, I would like to order a coke and fries please".

  3. Click the "Score Actions" button

    The user hasn't specified sizes for the drink or fries so we need to ask them.

  4. Select the action with response: "What size drink would you like?"

  5. As the user type in, "large"

  6. Click the "Score Actions" button

  7. Select the action SET_ENTITY - "drinkSize: LARGE"

  8. Select the action with response: "What size fries would you like?"

  9. As the use type in, "Um, make those a medium.

  10. Click the "Score Actions" button

  11. Select the action SET_ENTITY - "friesSize: MEDIUM"

  12. Select the action with response: "Would you like any condiments?"

  13. As the use type in, "Yes"

  14. Click the "Score Actions" button

  15. Select the action SET_ENTITY - "condimentsConfirmation: YES"

  16. Select the action with response: "Ok, I have an order for a LARGE drink and MEDIUM fries. Is that correct?"

  17. As the use type in, "Yes"

  18. Click the "Score Actions" button

  19. Select the action SET_ENTITY - "orderConfirmation: YES"

  20. Select the action with response: "Ok, your order number is 58. Please wait over there."

  21. Click "Save" to close the dialog

You have just created your first dialog using ENUM entities and SET_ENTITY actions. You can make many more combinations of the user specifying partial information or experimenting with other types of close-ended questions.

Note

During training you will see placeholders for the SET_ENTITY actions such as

action_set_entity_training.png

but when creating log dialogs or using deployed bots users will not see these.

Next steps

[!div class="nextstepaction"] Alternative inputs