Creating bots from raw logs and calculating the coverage of business logic graphs

Abstract

A method (500) for dynamically generating training data for a model involves receiving a transcript (152) corresponding to a conversation between a customer (12) and an agent (11), wherein the transcript includes a customer input (20) and an agent input (30). The method further involves receiving a logic model (300) containing a plurality of responses (310), each response of the plurality of responses representing a possible response to the customer input. The method further involves selecting a response (310) from the plurality of responses of the logic model based on the agent input.The procedure involves determining that a similarity value (253) between the selected response and the agent input satisfies a similarity threshold (255), and, based on determining that the similarity value between the selected response and the agent input satisfies the similarity threshold, training a machine learning model (210) using the customer input and the selected response.

Images