| CPC G06F 40/35 (2020.01) [G06F 40/237 (2020.01)] | 16 Claims |
|
1. A system comprising:
one or more processors configured to:
receive one or more user inputs; and
iteratively navigate a conversation graph according to the one or more user inputs, wherein the conversation graph comprises nodes and edges, each node representing a possible state of a conversation between an automated conversational agent and a user, each edge connecting two nodes thereby representing a transition between respective states of the conversation, wherein each node is associated with one or more respective predetermined actions, and wherein in each iteration of navigation of the conversation graph, the one or more processors are configured to:
process at least one user input of the one or more user inputs through a language model trained to receive the at least one user input and generate a function call based on the at least one user input, wherein the function call is an Application Programming Interface (API) call of an API;
determine a state change from a current node to a first next node that is next adjacent to the current node in the conversation graph according to the function call; and
perform the one or more predetermined actions associated with the first next node in the conversation graph,
wherein the one or more processors are further configured to:
train the language model until reaching one or more convergence criteria, wherein in training the language model, the one or more processors are configured to perform one or more iterations of:
sending, as input to the language model, a training example representing at least a portion of a session log labeled with an API call, the session log generated using the conversation graph, and
computing a loss between a generated output of the language model from the training example, with the labeled API call, and
updating one or more model parameter values of the language model based on the computed loss.
|