{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.09570", "kind": "paper", "url": "https://arxiv.org/abs/2606.09570"}, {"id": "gh:mengze-hong/uxbench", "kind": "repo", "url": "https://github.com/mengze-hong/UXBench"}], "authors": ["Mengze Hong", "Xia Zeng", "Zeyang Lei", "Sheng Wang", "Chen Jason Zhang", "Di Jiang", "Taiming Fu", "Jinfeng Huang", "Mengqiao Liu", "Qinghe Chang", "Haosheng Zou", "Qiongyi Zhou", "Sijun He", "Simonjmdeng", "Minlong Peng", "Di Liang"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.09570", "source_url": "https://arxiv.org/abs/2606.09570"}, "categories": ["General AI", "Language & Knowledge", "Safety"], "description": {"en": "UXBench evaluates user experience in AI assistants through three tasks: UX Judge (binary classification of response quality), UX Eval (response generation), and UX Recovery (repairing failed interactions). The dataset contains 7,400 test instances from 70K+ real interaction logs, covering 8 scenarios and 83 domains. Scoring uses accuracy for Judge and GRM-rated quality for generation tasks."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.09570", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.09570", "title": "UXBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2606.09570", "modality": null, "name": "UXBench", "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_uxbench_9ba76902", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "56f9db7266aec2dbcb5e4d07cd4e9d40d24e0d5adaeeba7daef0dded7fe92f2c", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.09570"}, "publisher": null, "released": "2026-06-08", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.09570", "source_url": "https://arxiv.org/abs/2606.09570"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-09570", "source": "claire_radar", "source_benchmark_id": "2606.09570", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Language & Knowledge", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/mengze-hong/UXBench", "githubScope": "benchmark_repo", "githubStars": 17, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 4, "hfPaperUrl": "https://huggingface.co/papers/2606.09570", "observedAt": "2026-10-04T15:31:48.030916Z", "signalStatus": {"githubStars": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-09-24", "state": "stale"}}, "status": "fresh"}, "capabilities": [], "confidence": "High", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "189775d557918107eabb50b66a7917c5b1edc6e09eed26da6dc9b5754ef0339c", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "UXBench evaluates user experience in AI assistants through three tasks: UX Judge (binary classification of response quality), UX Eval (response generation), and UX Recovery (repairing failed interactions). The dataset contains 7,400 test instances from 70K+ real interaction logs, covering 8 scenarios and 83 domains. Scoring uses accuracy for Judge and GRM-rated quality for generation tasks.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["coined title prefix ending in Bench or Benchmark", "benchmark release stated in one sentence", "evaluation protocol evidence"], "snippet": "We present UXBench, the first user-centric benchmark grounded in real user feedback signals for evaluating preference alignment and dialogue generation."}, "familyId": "bmf_8d180072bdf4", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [{"evidence": [{"basis": "existing-code-link", "linkedUrl": "https://github.com/mengze-hong/UXBench", "url": "https://github.com/mengze-hong/UXBench"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/mengze-hong/UXBench", "url": "https://huggingface.co/papers/2606.09570"}], "scope": "benchmark_repo", "url": "https://github.com/mengze-hong/UXBench", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_uxbench_9ba76902", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/mengze-hong/UXBench", "data": null, "hfPaper": "https://huggingface.co/papers/2606.09570", "paper": "https://arxiv.org/abs/2606.09570", "pdf": "https://arxiv.org/pdf/2606.09570", "project": null, "report": "https://arxiv.org/abs/2606.09570"}, "metrics": [], "motivation": "As AI assistants serve millions of users daily, evaluating user experience (UX) beyond general model capability has become increasingly important.", "name": "UXBench", "oneLine": "UXBench evaluates user experience in AI assistants through three tasks: UX Judge (binary classification of response quality), UX Eval (response generation), and UX Recovery (repairing failed interactions). The dataset contains 7,400 test instances from 70K+ real interaction logs, covering 8 scenarios and 83 domains. Scoring uses accuracy for Judge and GRM-rated quality for generation tasks.", "paperTitle": "UXBench: Benchmarking User Experience in AI Assistants", "primaryDomain": "General AI", "publication": {"evidence": "Accepted by EMNLP 2026 Main Conference", "evidenceLevel": "author-claim", "evidenceUrl": "https://arxiv.org/abs/2606.09570", "source": "arxiv-comments", "status": "acceptance_claimed", "venue": "EMNLP 2026 Main Conference", "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-08", "source": {"authors": ["Mengze Hong", "Xia Zeng", "Zeyang Lei", "Sheng Wang", "Chen Jason Zhang", "Di Jiang", "Taiming Fu", "Jinfeng Huang", "Mengqiao Liu", "Qinghe Chang", "Haosheng Zou", "Qiongyi Zhou", "Sijun He", "Simonjmdeng", "Haojing Huang", "Zijian Li", "Lucas Mu Li", "Fubao Zhang", "Mona Zhou", "Wei Ma", "Yuan Hua", "Qi Zhu", "Shuo Jiang", "Chenxuan Ma", "Yuanmeng Zhang", "Jian Song", "Minlong Peng", "Di Liang", "Davey Chen"], "categories": ["cs.CL", "cs.HC"], "id": "2606.09570", "title": "UXBench: Benchmarking User Experience in AI Assistants", "type": "arxiv", "url": "https://arxiv.org/abs/2606.09570"}, "sourceUpdatedAt": "2026-07-14T09:11:36Z", "topics": ["Safety"], "venueAttempts": [{"decisionRaw": "Accepted by EMNLP 2026 Main Conference", "evidence": [{"level": "author-claim", "observedAt": "2026-10-14T01:46:58.058199Z", "rawValue": "Accepted by EMNLP 2026 Main Conference", "sourceType": "arxiv-comments", "sourceUrl": "https://arxiv.org/abs/2606.09570"}], "reviewStatus": "accepted", "venueName": "EMNLP 2026 Main Conference"}], "whyItMatters": "UXBench addresses the gap in evaluating AI assistants beyond raw capability, focusing on user-perceived utility and preference alignment. It provides a structured way to measure how well models understand and improve user experience, offering practical value for developing assistants that better satisfy real users."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
