![[2602.19040] Adaptive Multi-Agent Reasoning for Text-to-Video Retrieval](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.19040] Adaptive Multi-Agent Reasoning for Text-to-Video Retrieval
Summary
The paper presents an adaptive multi-agent framework for improving text-to-video retrieval systems, addressing challenges in query-dependent temporal reasoning and achieving significant performance enhancements over existing methods.
Why It Matters
As short-form video content proliferates, effective retrieval systems are crucial for user engagement and content discovery. This research proposes a novel approach that enhances retrieval accuracy and efficiency, which is vital for applications in multimedia and AI-driven platforms.
Key Takeaways
- Introduces an adaptive multi-agent framework for text-to-video retrieval.
- Improves query-dependent temporal reasoning through specialized agents.
- Demonstrates a twofold performance improvement over existing methods.
- Utilizes a novel communication mechanism for better agent coordination.
- Achieves significant advancements on TRECVid benchmarks.
Computer Science > Information Retrieval arXiv:2602.19040 (cs) [Submitted on 2 Dec 2025] Title:Adaptive Multi-Agent Reasoning for Text-to-Video Retrieval Authors:Jiaxin Wu, Xiao-Yong Wei, Qing Li View a PDF of the paper titled Adaptive Multi-Agent Reasoning for Text-to-Video Retrieval, by Jiaxin Wu and Xiao-Yong Wei and Qing Li View PDF HTML (experimental) Abstract:The rise of short-form video platforms and the emergence of multimodal large language models (MLLMs) have amplified the need for scalable, effective, zero-shot text-to-video retrieval systems. While recent advances in large-scale pretraining have improved zero-shot cross-modal alignment, existing methods still struggle with query-dependent temporal reasoning, limiting their effectiveness on complex queries involving temporal, logical, or causal relationships. To address these limitations, we propose an adaptive multi-agent retrieval framework that dynamically orchestrates specialized agents over multiple reasoning iterations based on the demands of each query. The framework includes: (1) a retrieval agent for scalable retrieval over large video corpora, (2) a reasoning agent for zero-shot contextual temporal reasoning, and (3) a query reformulation agent for refining ambiguous queries and recovering performance for those that degrade over iterations. These agents are dynamically coordinated by an orchestration agent, which leverages intermediate feedback and reasoning outcomes to guide execution. We also introdu...
