Intents and Entities

Intent classification and Entity extraction

The dataset consists of training and testing phrases. These phrases contain important elements used to define the context of your AI, including intents and entities.

By using these elements, we have enough information to guide a conversation flow with a good level of detail.

The intelligence usage is done through intent classifications and entity extractions:

Intentions

Intents are what define the context of your intelligence. Upon receiving a sentence to classify, your intelligence will respond to which of the trained intents that sentence is most likely to belong to.

Therefore, it is crucial to make an early planning about the intelligence scope, so the intentions can be clearly and precisely defined.

By doing a detailed early planning, an eventual need of changing the intelligence's scope, removing or adding new intentions after launch is drastically reduced.

See here how to add intentions!

Entities

Entities are elements used to extract additional information when a phrase is sent to the bot, besides the classified intent.

The use of entities is an optional feature.

As the name suggests, entities are "things." In Weni, entities are defined by grouping words that belong to the same category, for example:

  • pepperoni, mozzarella, margherita as flavor
  • television, smartphone, tablet as product

Your AI will automatically try to search for flavors and products in the received phrases, extracting and returning them along with the classified intent.

It is important to note that the AI will also attempt to extract other words that were not explicitly marked by the user but are highly likely to belong to that word group (entity).

For instance, considering the product entity above, with just those three labels, your AI would also try to extract words like computer, notebook, and TV as products.

See here how to use entities!

The intent classification and entity extraction tasks do not interfere with each other, they are two independent processes and results.

How did we do?

Contact