基于知识图谱质量评估的大模型生成链路控制方法
By generating and executing structured query statements, and combining them with a large language model to generate answers, and employing symbolic and neural scoring mechanisms, this approach solves the problem of quantitative correlation between graph state changes and the support capability of generation links in knowledge graph quality assessment. This achieves unified and comparable assessment results and improves applicability under complex tasks.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SICHUAN UNIV JINCHENG INST
- Filing Date
- 2026-05-11
- Publication Date
- 2026-07-17
AI Technical Summary
Existing knowledge graph quality assessment methods struggle to establish a quantitative correlation between graph state changes and the supporting capabilities of generation links, lack a unified and comparable expression method, cannot independently characterize graph contributions, and have insufficient applicability of assessment results in complex task scenarios and under non-ideal conditions.
By generating and executing structured query statements, obtaining query results and query hit indicators, and combining them with a large language model to generate answers, extracting answer behavior features, and employing symbolic and neural scoring mechanisms to perform gating dual-path fusion, a comprehensive score is generated to control the generation process.
It achieves a quantitative correlation between changes in the state of the knowledge graph and the ability to support the generated tasks, forming a unified and comparable evaluation result, enhancing the applicability and interpretability under complex tasks and non-ideal conditions, and improving the auxiliary judgment ability of the evaluation result.
Smart Images
Figure CN122198149B_ABST