Method and system for intent revision and optimization of multi-turn dialogue context memory

Abstract

The present application relates to the technical field of artificial intelligence dialogue system, and particularly relates to a multi-round dialogue context memory intention correction and optimization method and system. The method jointly encodes user input and historical dialogue, extracts key semantic elements to construct an intention evolution relationship graph, performs context consistency verification on an initial intention recognition result, generates an intention correction candidate set when a conflict is detected, and dynamically adjusts historical dialogue context encoding weights based on the corrected intention recognition result. The present application effectively improves the accuracy of intention recognition and the context coherence in multi-round dialogue, and reduces the semantic drift risk.

Images