Glossary

Manu da Silva Updated by Manu da Silva

In this article, you will find the meaning of some common terms and expressions used in Weni. Most of these terms are found in NLU platforms, but some are specific to the Weni community, making it very important for all users to understand them.

  • Precision: A metric obtained from intelligence tests. It refers to the ratio of correctly predicted positive observations to the total predicted positive observations.
  • Algorithm: Generally, it is a block of code that performs a specific action given an input. In the context of NLU, the algorithm specifies how the training data will be processed to generate a model that will make user predictions.
  • Confidence: The certainty rate of your bot regarding its classification. It indicates how confident the intelligence is in interpreting the response, based on the trained dataset.
  • Dataset: A dataset is a collection of all the example phrases registered in Weni. The dataset "teaches" the bot what it should understand about each intent.
  • Entity: Represents a specific piece of information extracted from the user's input. Entities can be used to add additional levels of abstraction for a given context.
  • Test: The process used to validate your intelligence's dataset. In both Quick Test and the Tests tab, users can analyze whether the training has been effective based on certain metrics.
  • Interactions: A feature that gathers all phrases sent to the intelligence along with their prediction data. It is typically used to improve datasets based on these phrases, as it allows users to add these phrases to training and/or test datasets.
  • Integration (API): The way external services communicate with BotHub to use predictions. We can integrate these other platforms with BotHub by providing the intelligence's authorization data.
  • Intent: Represents the goal of a user's input. Usually, the user defines the intent for each type of response that the intelligence will need to predict.
  • NLP (Natural Language Processing): Technology that handles interactions between humans and computers by processing an input in natural language and interpreting it so that the machine can take actions based on that input.
  • Recall: A metric obtained from intelligence tests. It refers to the ratio of correctly predicted observations to all observations made in the test.
  • Training/Test Phrases: A training or test phrase is a simulation of an end-user input sentence. They are usually related to the main scope of the intelligence and are used to feed its datasets.
  • Training: The process of "teaching" new phrases from your dataset to improve intelligence. Running training means generating a new model (a new intelligence) for your bot based on all the registered training phrases.
  • Model: Essentially a set of weights generated by training. The training adjusts the model based on the dataset so that it learns the patterns and can thus answer questions it has never seen before (in our case, classify phrases). The model is the intelligence of the bot, and it performs the phrase classification. Weni abstracts the use of the model to make it simple for the user.
  • Versions: A feature that allows users to work with multiple versions of the same dataset, making changes without interference between them.

How did we do?

Interface Updates

Guidance and Best Practices

Contact