{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.25325", "kind": "paper", "url": "https://arxiv.org/abs/2606.25325"}, {"id": "gh:zhiyuanhan-aaron/oppo", "kind": "repo", "url": "https://github.com/ZhiyuanHan-Aaron/OPPO"}, {"kind": "website", "url": "https://zjycutieee.github.io/OPPO-page/"}], "authors": ["Zhiyuan Han", "Beier Zhu", "Wenwen Tong", "Pengyang Shao", "Peipei Song", "Xinyi Wang", "Jiangnan Chen", "Lewei Lu", "Xun Yang"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.25325", "source_url": "https://arxiv.org/abs/2606.25325"}, "categories": ["General AI", "Multimodal", "Reasoning"], "description": {"en": "MEP-Bench evaluates whether multimodal emotion models use modality-specific evidence sufficiently and faithfully."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.25325", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.25325", "title": "MEP-Bench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2606.25325", "modality": null, "name": "MEP-Bench", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"admission_policy_version": "2026-09-09.1", "confidence": "Medium", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-09-25T04:40:22Z", "data_status": "primary-source-indexed", "display_eligible": "true", "origin_record_id": "bm_mep-bench_d510909e", "origin_source": "arxiv", "recognition_confidence": "0.6", "record_sha256": "52f8c9dab10ebc278205d89907c4a2101e24f587e81fe86369aa3791885c8f95", "relation": "introduces", "review_model": "codex-cli-default", "review_state": "ai-reviewed", "reviewed_at": "2026-09-24T18:38:25.763173Z", "source_url": "https://arxiv.org/abs/2606.25325"}, "publisher": null, "released": "2026-06-24", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.25325", "source_url": "https://arxiv.org/abs/2606.25325"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-25325", "source": "claire_radar", "source_benchmark_id": "2606.25325", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Multimodal", "attentionForecast": {"confidence": "Medium", "horizon": "7d", "reason": "ICML acceptance, a public code link, and growing interest in multimodal reasoning and hallucination may generate near-term attention.", "score": 69}, "capabilities": ["Reasoning"], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-24T18:38:25.763173Z", "inputHash": "70913a00fda61518f878ccead857e221e11b153838bfbcd1b1aa221872a47496", "model": "codex-cli-default"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "We further introduce MEP-Bench", "canonicalNameSource": "abstract", "decisionReason": "The supplied evidence formally names the benchmark, identifies the evaluated properties and corresponding scores, and provides an official public code path supporting reuse.", "evaluationEvidence": {"reuse": "Code: https://github.com/ZhiyuanHan-Aaron/OPPO", "scoring": "utilization and faithfulness scores on MEP-Bench", "task": "a diagnostic benchmark that quantifies utilization and faithfulness"}, "model": "codex-cli-default", "reviewedAt": "2026-09-24T18:38:25.763173Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "MEP-Bench evaluates whether multimodal emotion models use modality-specific evidence sufficiently and faithfully.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["exact named benchmark artifact released in abstract", "public artifact URL"], "snippet": "We further introduce MEP-Bench, a diagnostic benchmark that quantifies utilization and faithfulness."}, "familyId": "bmf_770103af5900", "firstSeenAt": "2026-08-19", "heat": null, "id": "bm_mep-bench_d510909e", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/ZhiyuanHan-Aaron/OPPO", "data": null, "paper": "https://arxiv.org/abs/2606.25325", "pdf": "https://arxiv.org/pdf/2606.25325", "project": "https://zjycutieee.github.io/OPPO-page/", "report": "https://arxiv.org/abs/2606.25325"}, "metrics": [], "motivation": "We find that current emotion-oriented Omni-MLLMs still lack reliable omni-modal perception: they (i) underutilize multimodal cues in their reasoning trajectories and (ii) exhibit unfaithful behavior, often hallucinating modality-specific statements from other modalities.", "name": "MEP-Bench", "oneLine": "MEP-Bench evaluates whether multimodal emotion models use modality-specific evidence sufficiently and faithfully.", "paperTitle": "Omni-Perception Policy Optimization for Multimodal Emotion Reasoning", "primaryDomain": "General AI", "publication": {"evidence": "Accepted at ICML 2026. Project page: https://zjycutieee.github.io/OPPO-page/ Code: https://github.com/ZhiyuanHan-Aaron/OPPO", "evidenceLevel": "author-claim", "evidenceUrl": "https://arxiv.org/abs/2606.25325", "source": "arxiv-comments", "status": "acceptance_claimed", "venue": "ICML 2026", "verifiedAt": "2026-08-19T11:05:13.395391Z"}, "publications": [], "readiness": "Runnable", "recognitionConfidence": 0.6, "relation": "introduces", "releasedAt": "2026-06-24", "reviewLifecycle": {"state": "pending"}, "source": {"authors": ["Zhiyuan Han", "Beier Zhu", "Wenwen Tong", "Pengyang Shao", "Peipei Song", "Xinyi Wang", "Jiangnan Chen", "Lewei Lu", "and Xun Yang"], "categories": ["cs.AI"], "id": "2606.25325", "title": "Omni-Perception Policy Optimization for Multimodal Emotion Reasoning", "type": "arxiv", "url": "https://arxiv.org/abs/2606.25325"}, "sourceUpdatedAt": "2026-07-20T13:20:39Z", "topics": ["Multimodal", "Reasoning"], "venueAttempts": [{"decisionRaw": "Accepted at ICML 2026. Project page: https://zjycutieee.github.io/OPPO-page/ Code: https://github.com/ZhiyuanHan-Aaron/OPPO", "evidence": [{"level": "author-claim", "observedAt": "2026-08-19T11:05:13.395391Z", "rawValue": "Accepted at ICML 2026. Project page: https://zjycutieee.github.io/OPPO-page/ Code: https://github.com/ZhiyuanHan-Aaron/OPPO", "sourceType": "arxiv-comments", "sourceUrl": "https://arxiv.org/abs/2606.25325"}], "reviewStatus": "accepted", "venueName": "ICML 2026"}], "whyItMatters": "It separates multimodal cue utilization from faithfulness, exposing cross-modal hallucination that aggregate emotion performance may conceal."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
