一种基于蒙特卡洛树搜索进行指令增强的智能客服语言模型优化方法和系统
By generating a high-quality instruction set through Monte Carlo tree search and fine-tuning the intelligent customer service language model, the problems of comprehension bias and high resource requirements in complex scenarios of existing intelligent customer service systems are solved, and efficient complex scenario processing capabilities are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ZHEJIANG UNIV
- Filing Date
- 2024-03-06
- Publication Date
- 2026-07-17
AI Technical Summary
Existing intelligent customer service systems suffer from misunderstandings and biases when handling complex or personalized user commands, making it difficult to cope with question-and-answer interaction tasks in complex scenarios. Furthermore, traditional fine-tuning methods are costly and cannot guarantee the diversity and complexity of data.
A Monte Carlo tree search-based instruction enhancement method is adopted, which combines heuristic functions and reward mechanisms to generate a high-quality, diverse, and complex evolutionary instruction set. The search process is guided by a model scoring system, and the evolutionary instruction set is used to fine-tune the intelligent customer service language model.
It significantly improved the performance of the intelligent customer service system in complex dialogue and reasoning tasks, reduced the demand for computing resources, and enhanced the model's processing capabilities in complex scenarios.
Smart Images
Figure CN118170884B_ABST