MARQUIS: A Three-Stage Pipeline for Video Retrieval-Augmented Generation

2026-05-17

AI Summary

This work addresses the limitations of insufficient retrieval for complex queries and constrained cross-video higher-order reasoning in video retrieval-augmented generation. To overcome these challenges, the authors propose a novel three-stage framework: first enhancing retrieval relevance through query expansion and reranking; second, performing calibrated structured evidence extraction; and third, integrating the extracted evidence into controllable text generation, optionally guided by a retrieval-augmented language model (RLM). This approach is the first to jointly integrate query expansion, structured evidence extraction, and controllable generation, thereby surpassing the bottlenecks of single-embedding retrieval and long-context generation. On the MAGMaR2026 benchmark, the method substantially improves nDCG@10 from 0.195 to 0.759, raises the ITER-QA-BASE human evaluation score from 3.09 to 3.83, and achieves state-of-the-art citation recall performance among non-QA systems with MARQUIS-RLM.

Abstract

Retrieval-augmented generation from videos requires systems to retrieve relevant audiovisual evidence from large corpora and synthesize it into coherent, attributed text. Current approaches struggle at both ends: retrieval methods fail on complex, multi-faceted queries that cannot be captured by a single embedding, while generation methods lack the high-level reasoning needed to synthesize across multiple videos and face memory constraints over long, multi-video contexts. We present MARQUIS: a three-stage pipeline that addresses these limitations through (1) query expansion, fusion, and reranking, (2) calibrated structured evidence extraction, and (3) article generation from extracted evidence, optionally controlled by an RLM. On the MAGMaR2026 shared task, we improve retrieval performance from 0.195 to 0.759 (nDCG@10). For article generation, ITER-QA-BASE improves average human score from 3.09 to 3.83 over the CAG baseline, while MARQUIS-RLM achieves a human score of 3.30 and the strongest citation recall among non-QA systems.

Problem

• video retrieval-augmented generation
• complex queries
• multi-video synthesis
• long-context reasoning
• audiovisual evidence retrieval

Innovation

• retrieval-augmented generation
• video understanding
• query expansion
• structured evidence extraction
• retrieval language model