Updated by Thays Oliveira
In this article, you can find the meaning of some terms and expressions that you may found in BotHub. Most of these terms are largely used in NLP platforms but a few of them are specific parts of the BotHub community and it is very important that all users are familiar with them.
- Accuracy/Precision: is the ratio of correctly predicted positive observations to the total predicted positive observations;
- Algorithm: is a block code that performs determined action given an input. For example, BotHub's algorithm specifies the process of choosing and training the model that will be used to classify user input;
- Confidence: is the model's certainty rate of its classification task. It means how sure of the interpretation the intelligence is in your answer, based on the trained dataset.
- Datasets: a dataset is the set of all training sentences (data) given to BotHub so it can better predict other sentences;
- Entity: represents a specific piece of information in the user's input. It can be used as another level of abstraction for determined context;
- Evaluate: is the process of testing your dataset so you can analyze the Accuracy and Recall ratios and improve your dataset;
- Inbox: is the feature that gathers all the inputs from the user that the intelligence has received. It can be used to improve your datasets with real data from the final users;
- Integration (API): is the form that BotHub communicates with other services to use its prediction features. We can integrate with other platforms just by giving the authorization token for each intelligence;
- Intent: represents the purpose of a user's input. Usually, you define an intent for each type of user request you want your application to predict;
- Labels: a label is a keyword used to categorize important entities on texts;
- NLP: NLP (Natural Language Processing) is the technology that takes care of interactions between humans and computers by processing the Natural Language input given by the human and interpreting it in many ways;
- Recall: is the ratio of correctly predicted positive observations to all observations in the test;
- Sentence: a sentence (or utterance) is a phrase related to a given subject that simulates a user's entry on the bot. A lot of sentences are required for the bot to start to build an interpretation pattern and predict better-related inputs from the user;
- Training: is the process of "teaching" new sentences from your dataset to improve the model's intelligence. Training your intelligence means generating a new model (a new intelligence) for your bot based on all registered training phrases.
- Model: is essentially a set of weights generated by training. The training adjusts the model based on the dataset so that it learns its patterns, thus being able to answer questions never seen by it before (in our case, classify sentences).
- Translation: is the process of translation existing datasets that are training in the intelligence;
- Versions: is the feature that allows the user to work in many versions of the same dataset without any interference between them;