NLU classifier algorithms
The following algorithms are available in JAICP:
-
STS (Semantic Textual Similarity) is an algorithm that compares the semantic distance between words. It considers inversion, dictionary forms of words, their synonyms, and other information.
-
Classic ML is a standard machine-learning algorithm for intent recognition based on logistic regression. It classifies text by original words, dictionary forms, and word stems. This algorithm does not consider semantic information.
-
Deep Learning is an algorithm based on convolutional neural networks. It considers the semantics of words when forming hypotheses.
-
Transformer ru and Transformer multi are multilingual algorithms. They evaluate the semantic similarity between the client request and all training phrases in the intent.
infoSelect the Transformer algorithm in the project settings:
- If NLU language is set to Russian, the system uses Transformer ru.
- For all other languages, it uses Transformer multi.
To choose the optimal algorithm for your project, refer to our algorithm ranking and detailed comparison. You can also learn about solutions for common algorithm issues.
In the project settings, you can select an algorithm and configure the classifier parameters.