Cross-system agent calling method based on natural language multi-turn dialogue

Abstract

This invention discloses a cross-system intelligent agent invocation method based on multi-turn natural language dialogue, belonging to the field of intelligent agent invocation technology. The method includes the following steps: parsing the user's multi-turn natural language task, generating a cross-system invocation chain, and determining the acceptance conditions of each invocation node in the chain; executing the current invocation node and generating an acceptance credential based on its return result; matching the acceptance credential with the acceptance conditions of the next invocation node, generating the next invocation parameters based on the matching result, and executing the next invocation node; repeatedly executing the steps of generating acceptance credentials and matching execution until the invocation chain is completed; responding to a correction action executed by the user, based on the parameter dependencies in the generated acceptance credentials, partially rolling back and resuming execution of the affected unexecuted invocation nodes. By controlling cross-node parameter reuse through acceptance credentials and parameter dependencies, the method solves the problem of incorrect transmission of corrected parameters, object change results, or results of inconsistent previous states to subsequent nodes.

Images