Brief

Researchers have developed FLOWREADER, a novel method for question answering over long, multimodal documents. This approach reframes evidence assembly as a min-cost flow problem, enabling better handling of fragmented information across text, tables, and slides. FLOWREADER outperforms traditional top-k retrieval methods on specific subsets of the VisDoMBench benchmark, demonstrating its effectiveness in complex evidence assembly scenarios. AI

Researchers have developed a new retrieval method called Constrained Dominant Sets (CDS) for multimodal document question answering. This technique addresses limitations in current systems that struggle with long documents by selecting complementary evidence rather than near-duplicates. CDS encodes the query as a structural constraint, automatically balances relevance and redundancy, and avoids greedy heuristics by achieving global equilibrium. When used with a Qwen3-VL-32B reader, CDS sets a new state-of-the-art on VisDoMBench and significantly improves performance on MMLongBench-Doc. AI