Predictive service scaling for conversational computing

Abstract

From intent data of a conversational system, a set of intent sequences and a model predicting a next intent for an intent sequence are constructed. A first intent is received as an input. Using the model, a next intent corresponding to the first intent is predicted. A service required by the next intent is determined. A resource consumption of the service is forecasted. Responsive to the forecasted resource consumption exceeding a present resource allocation to the service, it is concluded that the service requires upscaling before becoming available for use by the next intent. An availability time by which the service is required to be available for use by the next intent is determined. An initial time at which upscaling must begin to ensure that the service is available at the availability time is determined. Upscaling of the service is caused to be scheduled for the initial time.