Multi-round dialogue context memory intention correction and optimization method and system
By optimizing the context memory of a multi-turn dialogue system through joint semantic encoding and intent evolution relationship graph, the problems of intent recognition errors and inconsistent interactions in existing technologies are solved, and the dialogue system achieves high accuracy and consistency.
Patent Information
- Application Number
- CN202610064396.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-19
- Publication Date
- 2026-02-17
- Estimated Expiration
- 2046-01-19
AI Technical Summary
Existing multi-turn dialogue systems cannot dynamically adjust the importance of historical turns in their context memory mechanism design, leading to the introduction of irrelevant context noise or the omission of key information. Furthermore, they lack systematic intent evolution modeling, making it difficult to capture user intent transfer patterns, resulting in intent recognition errors and inconsistent interactions.
By jointly generating semantic vector representations of the fusion context through semantic encoding, constructing an intent evolution relationship graph, performing consistency verification on the initial intent recognition results, dynamically adjusting encoding weights, generating corrected intent recognition results, and optimizing the contextual understanding of the dialogue system.
It improves the accuracy of intent recognition and the coherence of interaction in multi-turn dialogues, can correct intent recognition biases in real time, optimize the system's contextual understanding capabilities, and maintain high accuracy and coherence in long dialogue interactions.