How to retrain an NLP model without going crazy

Issue — the constant retraining of classification models used by virtual assistants (VAs) can dramatically alter responses to user queries.

Risks — this can lead to unpredictable and unsatisfactory outcomes, negatively impacting the user experience.

Solution — development and implementation of an automated analytical system that enables the effective evaluation of the quality of new models and informed decisions about their deployment in a production environment. The solution aims to optimize the modelling process and ensure stable and predictable results for VA users.