Sparse Evidence Can Suffice: Agentic Evidence Seeking for Multimodal Video Misinformation Detection

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Haochen Zhao, Yongxiu Xu, Xinkui Lin, Dong Xie, Jiarui Lu, Yuqi Qian, Yubin Wang, Hongbo Xu, Gaopeng Gou
• Venue: arXiv: cs.CV
• Published: 2026-07-20
• Collected: 2026-07-21
• URL: http://arxiv.org/abs/2607.18080v1
• PDF: https://arxiv.org/pdf/2607.18080v1
• DOI: UNKNOWN

Plain-Language Brief

The paper proposes a new framework called SIEVE for detecting video misinformation. Instead of processing the entire video at once, this method focuses on identifying sparse, decision-relevant clues first and then verifying the veracity based on these clues. The authors train an evidence-seeking agent to explore multimodal evidence and construct a compact evidence package for verification. This approach is shown to outperform baselines in multiple video misinformation benchmarks and improves transparency by providing an explicit and inspectable evidence trail.

What the Document Says

• The paper presents SIEVE, a framework for sparse interactive evidence verification via extraction in multimodal video misinformation detection.
• The framework involves an evidence-seeking agent that explores multimodal evidence and constructs a compact evidence package for verification.
• The agent is trained with supervised evidence-seeking trajectories and an evidence-aware reinforcement learning objective.
• SIEVE outperforms the evaluated baselines in multiple video misinformation benchmarks.
• SIEVE supports reliable verification using compact evidence packages.
• The resulting acquisition process provides an explicit and inspectable evidence trail.
• This improves transparency and groundedness of multimodal misinformation detection.

Key Concepts Mentioned

• multimodal video misinformation detection
• SIEVE framework
• evidence-seeking agent
• compact evidence package
• supervised learning
• reinforcement learning
• video misinformation benchmarks
• transparency
• groundedness

Claims Made by the Document

• SIEVE outperforms the evaluated baselines.
• SIEVE supports reliable verification using compact evidence packages.

Evidence Used

• Experiments on multiple video misinformation benchmarks show that SIEVE consistently outperforms the evaluated baselines and supports reliable verification using compact evidence packages.

Methods / Approach

• Evidence-seeking agent
• Supervised learning
• Reinforcement learning

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

The paper does not mention any limitations of the proposed SIEVE framework.

Open Questions

• What are the specific video misinformation benchmarks used in the experiments?
• How is the performance of SIEVE compared to state-of-the-art methods?

Notes for Future Q&A

May include interpretation. Labeled where it does.

The text does not provide enough information about the specific video misinformation benchmarks used or a comparison with state-of-the-art methods.

Raw Abstract / Excerpt

Multimodal video misinformation detection is commonly formulated as a holistic video-understanding task, where the entire video and its associated content are processed and judged in a single pass. However, real-world misinformation often exhibits a sparse and compositional evidence structure: a reliable decision may depend on only a few coupled clues, while most video content contributes limited additional information. Exhaustive multimodal reasoning may therefore introduce substantial redundancy and obscure decisive evidence. This motivates decoupling evidence acquisition from verification: first identifying sparse, decision-relevant clues and then judging veracity based on the acquired evidence. Accordingly, we propose SIEVE, a framework for Sparse Interactive Evidence Verification via Extraction in multimodal video misinformation detection. An evidence-seeking agent actively explores the available multimodal evidence and constructs a compact evidence package, which is then used by a verifier to determine veracity. The agent is trained with supervised evidence-seeking trajectories and an evidence-aware reinforcement learning objective that promotes informative evidence acquisition while discouraging unnecessary or invalid interactions. Experiments on multiple video misinformation benchmarks show that SIEVE consistently outperforms the evaluated baselines and supports reliable verification using compact evidence packages. Moreover, the resulting acquisition process provides an explicit and inspectable evidence trail, improving the transparency and groundedness of multimodal misinformation detection.