{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:XQMNBUGI5UAXOD2UL3ABKYDLOS","short_pith_number":"pith:XQMNBUGI","schema_version":"1.0","canonical_sha256":"bc18d0d0c8ed01770f545ec015606b74982de6bd87ad93611b7baee3c7fb5bf1","source":{"kind":"arxiv","id":"2506.21600","version":1},"attestation_state":"computed","paper":{"title":"Structured Attention Matters to Multimodal LLMs in Document Understanding","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.IR"],"primary_cat":"cs.CL","authors_text":"Chang Liu, Hang Wu, Hongkai Chen, Ming-Hsuan Yang, Qingwen Ye, Yiwei Wang, Yujun Cai","submitted_at":"2025-06-19T07:16:18Z","abstract_excerpt":"Document understanding remains a significant challenge for multimodal large language models (MLLMs). While previous research has primarily focused on locating evidence pages through precise multimodal queries, our work investigates a fundamental yet overlooked aspect: how input format influences document comprehension performance. Through systematic analysis, we discover that raw OCR text often impairs rather than improves MLLMs' performance, which is a counterintuitive finding we attribute to attention dispersion and structure loss. To further substantiate our hypothesis, we propose a novel s"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2506.21600","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-06-19T07:16:18Z","cross_cats_sorted":["cs.AI","cs.IR"],"title_canon_sha256":"bfe72401d30daf93931042f5245664081269235243f26f1116d225793257720c","abstract_canon_sha256":"d4cd4622a64dfde23702b18c607221f6b07e54ec66f6174846c8508d8ec4b220"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:27:57.052952Z","signature_b64":"/tfSI/mIUhMo5YcW2nLaoediqQqAU4eH2JYu/IAfjPIgqugV7kZehlaHbF7H+MmoAst9p0E1BYvQv+yTJcWWCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bc18d0d0c8ed01770f545ec015606b74982de6bd87ad93611b7baee3c7fb5bf1","last_reissued_at":"2026-07-05T11:27:57.052512Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:27:57.052512Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Structured Attention Matters to Multimodal LLMs in Document Understanding","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.IR"],"primary_cat":"cs.CL","authors_text":"Chang Liu, Hang Wu, Hongkai Chen, Ming-Hsuan Yang, Qingwen Ye, Yiwei Wang, Yujun Cai","submitted_at":"2025-06-19T07:16:18Z","abstract_excerpt":"Document understanding remains a significant challenge for multimodal large language models (MLLMs). While previous research has primarily focused on locating evidence pages through precise multimodal queries, our work investigates a fundamental yet overlooked aspect: how input format influences document comprehension performance. Through systematic analysis, we discover that raw OCR text often impairs rather than improves MLLMs' performance, which is a counterintuitive finding we attribute to attention dispersion and structure loss. To further substantiate our hypothesis, we propose a novel s"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.21600","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2506.21600/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2506.21600","created_at":"2026-07-05T11:27:57.052568+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.21600v1","created_at":"2026-07-05T11:27:57.052568+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.21600","created_at":"2026-07-05T11:27:57.052568+00:00"},{"alias_kind":"pith_short_12","alias_value":"XQMNBUGI5UAX","created_at":"2026-07-05T11:27:57.052568+00:00"},{"alias_kind":"pith_short_16","alias_value":"XQMNBUGI5UAXOD2U","created_at":"2026-07-05T11:27:57.052568+00:00"},{"alias_kind":"pith_short_8","alias_value":"XQMNBUGI","created_at":"2026-07-05T11:27:57.052568+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.15223","citing_title":"GenAI-Driven Approach to RISC-V Supply Chain Exploration","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2604.05418","citing_title":"VideoStir: Understanding Long Videos via Spatio-Temporally Structured and Intent-Aware RAG","ref_index":93,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XQMNBUGI5UAXOD2UL3ABKYDLOS","json":"https://pith.science/pith/XQMNBUGI5UAXOD2UL3ABKYDLOS.json","graph_json":"https://pith.science/api/pith-number/XQMNBUGI5UAXOD2UL3ABKYDLOS/graph.json","events_json":"https://pith.science/api/pith-number/XQMNBUGI5UAXOD2UL3ABKYDLOS/events.json","paper":"https://pith.science/paper/XQMNBUGI"},"agent_actions":{"view_html":"https://pith.science/pith/XQMNBUGI5UAXOD2UL3ABKYDLOS","download_json":"https://pith.science/pith/XQMNBUGI5UAXOD2UL3ABKYDLOS.json","view_paper":"https://pith.science/paper/XQMNBUGI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.21600&json=true","fetch_graph":"https://pith.science/api/pith-number/XQMNBUGI5UAXOD2UL3ABKYDLOS/graph.json","fetch_events":"https://pith.science/api/pith-number/XQMNBUGI5UAXOD2UL3ABKYDLOS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XQMNBUGI5UAXOD2UL3ABKYDLOS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XQMNBUGI5UAXOD2UL3ABKYDLOS/action/storage_attestation","attest_author":"https://pith.science/pith/XQMNBUGI5UAXOD2UL3ABKYDLOS/action/author_attestation","sign_citation":"https://pith.science/pith/XQMNBUGI5UAXOD2UL3ABKYDLOS/action/citation_signature","submit_replication":"https://pith.science/pith/XQMNBUGI5UAXOD2UL3ABKYDLOS/action/replication_record"}},"created_at":"2026-07-05T11:27:57.052568+00:00","updated_at":"2026-07-05T11:27:57.052568+00:00"}