{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2605.31349", "kind": "paper", "url": "https://arxiv.org/abs/2605.31349"}], "authors": ["Paramananda Bhaskar", "Naquee Rizwan", "Daksh Jogchand", "Saurabh Kumar Pandey", "Animesh Mukherjee"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2605.31349", "source_url": "https://arxiv.org/abs/2605.31349"}, "categories": ["Cybersecurity", "Vision & 3D", "cs.CL"], "description": {"en": "FBHM evaluates vision-language models on hateful meme detection across 25 rhetorical functionalities and 10 target communities, with 5,000 memes. Performance is measured by Macro-F1 score on this curated dataset."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2605.31349", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2605.31349", "title": "FBHM"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2605.31349", "modality": null, "name": "FBHM", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"confidence": "High", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-10-14T09:44:51Z", "data_status": "primary-source-indexed", "display_eligible": "unknown", "origin_record_id": "bm_fbhm_44ffb389", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "585d85a0554701c6bf010e462becbc2ea0b8abaed698bccb254a7e463dc3fc04", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2605.31349"}, "publisher": null, "released": "2026-05-29", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2605.31349", "source_url": "https://arxiv.org/abs/2605.31349"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2605-31349", "source": "claire_radar", "source_benchmark_id": "2605.31349", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["Cybersecurity"], "area": "Vision & 3D", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": null, "githubScope": null, "githubStars": null, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 1, "hfPaperUrl": "https://huggingface.co/papers/2605.31349", "observedAt": "2026-10-04T03:02:42.754940Z", "signalStatus": {"githubStars": {"state": "not_applicable"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-09-30", "state": "stale"}}, "status": "stale"}, "capabilities": [], "confidence": "High", "construction": "Transform Existing", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "65fa90f444451fd25fa576d244646aea002efee6a85f595e49f3dc0616e67931", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "FBHM evaluates vision-language models on hateful meme detection across 25 rhetorical functionalities and 10 target communities, with 5,000 memes. Performance is measured by Macro-F1 score on this curated dataset.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact coined title identity tied to benchmark evidence", "benchmark release stated in one sentence", "evaluation protocol evidence"], "snippet": "To address this, we introduce FBHM, a systematically curated benchmark of Functionality Based Hateful Memes constructed along two orthogonal axes: 25 distinct rhetorical functionalities and 10 target communities (5,000 memes total)."}, "familyId": "bmf_752147a6df00", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_fbhm_44ffb389", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": ["Cybersecurity"], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2605.31349", "paper": "https://arxiv.org/abs/2605.31349", "pdf": "https://arxiv.org/pdf/2605.31349", "project": null, "report": "https://arxiv.org/abs/2605.31349"}, "metrics": [], "motivation": "Hateful meme detection remains a formidable challenge for vision-language models, as existing benchmarks are structurally observational - confounding rhetorical hate mechanisms with target community features and preventing causal evaluation of model vulnerabilities.", "name": "FBHM", "oneLine": "FBHM evaluates vision-language models on hateful meme detection across 25 rhetorical functionalities and 10 target communities, with 5,000 memes. Performance is measured by Macro-F1 score on this curated dataset.", "paperTitle": "FBHM: Functional Benchmarking and Steering of VLMs for Hateful Meme Detection", "primaryDomain": "Cybersecurity", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2605.31349", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "arXiv", "organizationType": "community", "role": "benchmark-publisher", "sourceUrl": "https://arxiv.org/abs/2605.31349"}], "ranking": {}, "readiness": "Paper only", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-05-29", "source": {"authors": ["Paramananda Bhaskar", "Naquee Rizwan", "Daksh Jogchand", "Saurabh Kumar Pandey", "Animesh Mukherjee"], "categories": ["cs.CL", "cs.AI", "cs.CV", "cs.MM"], "id": "2605.31349", "title": "FBHM: Functional Benchmarking and Steering of VLMs for Hateful Meme Detection", "type": "arxiv", "url": "https://arxiv.org/abs/2605.31349"}, "sourceUpdatedAt": "2026-05-29T14:27:17Z", "topics": ["cs.CL"], "venueAttempts": [], "whyItMatters": "Existing hateful meme benchmarks confound rhetorical strategies with target community features, preventing causal evaluation of model vulnerabilities. FBHM isolates these axes, revealing that models rely on dataset-specific heuristics rather than robust reasoning, and offers a controlled environment for measuring generalization."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
