BRIDGE: Benchmark for multi-hop Reasoning In long multimodal Documents with Grounded Evidence
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , |
|---|---|
| Format: | Preprint |
| Publié: |
2026
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866915845670699008 |
|---|---|
| author | Xiang, Biao Han, Soyeon Caren Ding, Yihao |
| author_facet | Xiang, Biao Han, Soyeon Caren Ding, Yihao |
| contents | Multi-hop question answering (QA) is widely used to evaluate the reasoning capabilities of large language models, yet most benchmarks focus on final answer correctness and overlook intermediate reasoning, especially in long multimodal documents. We introduce BRIDGE, a benchmark for multi-hop reasoning over long scientific papers that require integrating evidence across text, tables, and figures. The dataset supports both chain-like and fan-out structures and provides explicit multi-hop reasoning annotations for step-level evaluation beyond answer accuracy. Experiments with state-of-the-art LLMs and multimodal retrieval-augmented generation (RAG) systems reveal systematic deficiencies in evidence aggregation and grounding that remain hidden under conventional answer-only evaluation. BRIDGE provides a targeted testbed for diagnosing reasoning failures in long multimodal documents. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_07931 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | BRIDGE: Benchmark for multi-hop Reasoning In long multimodal Documents with Grounded Evidence Xiang, Biao Han, Soyeon Caren Ding, Yihao Computation and Language Multi-hop question answering (QA) is widely used to evaluate the reasoning capabilities of large language models, yet most benchmarks focus on final answer correctness and overlook intermediate reasoning, especially in long multimodal documents. We introduce BRIDGE, a benchmark for multi-hop reasoning over long scientific papers that require integrating evidence across text, tables, and figures. The dataset supports both chain-like and fan-out structures and provides explicit multi-hop reasoning annotations for step-level evaluation beyond answer accuracy. Experiments with state-of-the-art LLMs and multimodal retrieval-augmented generation (RAG) systems reveal systematic deficiencies in evidence aggregation and grounding that remain hidden under conventional answer-only evaluation. BRIDGE provides a targeted testbed for diagnosing reasoning failures in long multimodal documents. |
| title | BRIDGE: Benchmark for multi-hop Reasoning In long multimodal Documents with Grounded Evidence |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2603.07931 |