公开(公告)号：EP3905149A1

公开(公告)日：2021-11-03

申请号：EP21170869.8

申请日：2021-04-28

申请人： Directly, Inc.

发明人： Todd, Jason ,  Ozdemir, Sinan ,  Abdullah, Shayaan Ahmad

IPC分类号： G06Q10/00 ,  G06Q10/02 ,  G06Q10/10 ,  G06Q50/10 ,  G06Q50/18

摘要： Issues relating to the setup and maintenance of virtual agents have been, to a substantial degree, resolved through the use of a dynamically adaptable knowledge graph. The knowledge graph is rapidly produced by organizing conversation statements into a multi-level hierarchy of clusters. Each of these clusters is associated with a node of the knowledge graph, and edges between the nodes represent the orders in which statements may be made in a conversation with a virtual agent. Maintenance and refinement of the knowledge graph is greatly simplified by monitoring virtual agent performance at a node specific level and employing human agents to moderate changes at individual nodes. With this approach, the performance of a virtual agent can be constantly improved by targeted human intervention. The virtual agent can even adapt in real-time to changing conditions and needs. For example, nodes of the knowledge graph may be quickly and automatically adapted to the introduction of a new product or discovery of a new customer issue. As is discussed further herein, a substantial improvement in the generation, maintenance and performance of virtual agents is achieved through a judicious and groundbreaking combination of machine learning and human augmentation.