---
id: 20260327-T0-14
title: "SCoOP：降低多模态模型幻觉风险的语义一致性方法"
title_en: "SCoOP Reduces Hallucinations in Multi-Model Systems"
url: https://ai.daily.yangsir.net/daily/20260327-T0-14
issue_date: 2026-03-27
publish_date: 2026-03-26T04:00:00.000Z
category: research
source_name: "arXiv cs.AI"
source_url: https://arxiv.org/abs/2603.23853
---

# SCoOP：降低多模态模型幻觉风险的语义一致性方法

研究团队提出SCoOP方法，解决多视觉语言模型集成时的不确定性问题。当多个异构模型协同工作时，输出聚合会放大不确定性并增加幻觉风险。SCoOP通过语义一致性约束，将多个模型的预测结果进行整合，实验表明能有效降低25%的幻觉发生率，为构建更稳健的多模态AI系统提供技术支持。

## English Version

**SCoOP Reduces Hallucinations in Multi-Model Systems**

Researchers introduce SCoOP to address uncertainty in combining multiple Vision-Language Models (VLMs). Aggregating heterogeneous model outputs amplifies hallucination risks. SCoOP uses semantic consistency constraints to integrate predictions, reducing hallucination rates by 25% in experiments, providing technical support for more robust multimodal AI systems.

---

**来源**：[arXiv cs.AI](https://arxiv.org/abs/2603.23853)

**详情页**：https://ai.daily.yangsir.net/daily/20260327-T0-14

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*