Intent Classification for a Management Conversational Assistant

Abstract

Intent classification is an essential step in processing user input to a conversational assistant. This work investigates techniques of intent classification of chat messages used for communication among software development teams with the aim of building an intent classifier for a management conversational assistant integrated into modern communication platforms used by developers. Experiments conducted using rule-based and common ML techniques have shown that careful choice of classification features has a significant impact on performance, and the best performing model was able to obtain a classification accuracy of 72%. A set of techniques for extracting useful features for text classification in the software engineering domain was also implemented and tested.