Knowledge Learning for Cognitive Business Conversations

Abstract

Cognitive conversation services are increasingly popular among lots of business companies. Cognitive conversation services enable a business company to expose its business functionalities directly to its customers in a user-friendly conversational mode, usually in the format of procedure dialog. The main challenge, however, is the impractical and insufficient creation process to manually build all such procedure dialogs. It also remains unclear how to optimize such procedure dialogs. In this paper, we propose a framework for incrementally mining procedure dialogs from business conversations. Our framework takes the procedure dialogs as initial input to generate machine learning models, then incorporates runtime user interactions to update the model using reinforcement learning, and finally transforms the refined model into the updates on existing procedure dialogs (or derive new dialog candidates) in a human-readable format so that Subject Matter Experts (SMEs) can understand and intervene in the further improvement process.