A knowledge-enhanced traceable multi-modal reasoning method, device and medium
By constructing structured candidate hypotheses and evidence-constrained reasoning, the problem of unclear evidence support in multimodal reasoning methods is solved, enabling explicit verification of high-confidence evidence and reliable generation of answers, thereby improving the interpretability and accuracy of multimodal reasoning.
Patent Information
- Application Number
- CN202610837530.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-11
- Publication Date
- 2026-07-17
AI Technical Summary
Existing multimodal reasoning methods suffer from problems in interpretability and reliability, including unaddressable evidence in attention methods, irrelevant noise knowledge in knowledge enhancement methods, lack of cross-modal consistency in multi-stage chained reasoning, and separation of semantic verification in visual localization, which leads to difficulty in obtaining clear evidence support and instability in answer generation.
We employ a knowledge-enhanced traceable multimodal reasoning method, which constructs structured candidate hypotheses through hypothesis-driven knowledge discovery, dual-channel evidence verification, and evidence-constrained reasoning stages. We then conduct fine-grained evidence verification and answer decision-making, and use high-confidence evidence to constrain the final reasoning process, resulting in explicit verification scores and hierarchical prompts.
It significantly improves the interpretability, verifiability, and traceability of the reasoning process, reduces interference from irrelevant concepts, enhances the accuracy of evidence screening and the reliability of answer decisions, reduces the risk of error accumulation, and strengthens the stability of multimodal reasoning.
Smart Images
Figure CN122414409A_ABST