| { |
| "@context": { |
| "@language": "en", |
| "@vocab": "https://schema.org/", |
| "arrayShape": "cr:arrayShape", |
| "citeAs": "cr:citeAs", |
| "column": "cr:column", |
| "conformsTo": "dct:conformsTo", |
| "containedIn": "cr:containedIn", |
| "cr": "http://mlcommons.org/croissant/", |
| "data": { |
| "@id": "cr:data", |
| "@type": "@json" |
| }, |
| "dataBiases": "cr:dataBiases", |
| "dataCollection": "cr:dataCollection", |
| "dataType": { |
| "@id": "cr:dataType", |
| "@type": "@vocab" |
| }, |
| "dct": "http://purl.org/dc/terms/", |
| "extract": "cr:extract", |
| "field": "cr:field", |
| "fileProperty": "cr:fileProperty", |
| "fileObject": "cr:fileObject", |
| "fileSet": "cr:fileSet", |
| "format": "cr:format", |
| "includes": "cr:includes", |
| "isArray": "cr:isArray", |
| "isLiveDataset": "cr:isLiveDataset", |
| "jsonPath": "cr:jsonPath", |
| "key": "cr:key", |
| "md5": "cr:md5", |
| "parentField": "cr:parentField", |
| "path": "cr:path", |
| "personalSensitiveInformation": "cr:personalSensitiveInformation", |
| "recordSet": "cr:recordSet", |
| "references": "cr:references", |
| "regex": "cr:regex", |
| "repeated": "cr:repeated", |
| "replace": "cr:replace", |
| "sc": "https://schema.org/", |
| "separator": "cr:separator", |
| "source": "cr:source", |
| "subField": "cr:subField", |
| "transform": "cr:transform", |
| "rai": "http://mlcommons.org/croissant/RAI/", |
| "prov": "http://www.w3.org/ns/prov#" |
| }, |
| "@type": "sc:Dataset", |
| "distribution": [ |
| { |
| "@type": "cr:FileObject", |
| "@id": "repo", |
| "name": "repo", |
| "description": "The Hugging Face git repository.", |
| "contentUrl": "https://huggingface.co/datasets/PrivacySIM/PrivacySIM/tree/refs%2Fconvert%2Fparquet", |
| "encodingFormat": "git+https", |
| "sha256": "https://github.com/mlcommons/croissant/issues/80" |
| } |
| ], |
| "recordSet": [], |
| "conformsTo": "http://mlcommons.org/croissant/1.1", |
| "name": "PrivacySIM", |
| "description": "\n\t\n\t\t\n\t\tPrivacySIM (anonymous review version)\n\t\n\n\nAnonymous double-blind submission. This dataset card has been redacted\nto remove author and institution information for the review period. Full\nattribution, citation, code link, and DOI will be added after acceptance.\n\n\n\t\n\t\t\n\t\tSummary\n\t\n\nPrivacySIM aggregates user privacy preferences from five published user\nstudies into a unified evaluation suite for LLM simulation of user privacy behavior. Each row represents one participant's responses to a… See the full description on the dataset page: https://huggingface.co/datasets/PrivacySIM/PrivacySIM.", |
| "alternateName": [ |
| "PrivacySIM/PrivacySIM" |
| ], |
| "creator": { |
| "@type": "Person", |
| "name": "PrivacySIM", |
| "url": "https://huggingface.co/PrivacySIM" |
| }, |
| "keywords": [ |
| "English", |
| "cc-by-4.0", |
| "1K<n<10K", |
| "🇺🇸 Region: US", |
| "privacy", |
| "llm-evaluation", |
| "contextual-integrity", |
| "privacy-norms", |
| "user-study", |
| "alignment" |
| ], |
| "license": "https://choosealicense.com/licenses/cc-by-4.0/", |
| "url": "https://huggingface.co/datasets/PrivacySIM/PrivacySIM", |
| "rai:dataLimitations": "Some limitations involve: (1) Limited coverage in diverse populations Participants are predominantly English-speaking, with one Mandarin-Chinese study, and skew toward Western, European, and East-Asian populations. Hence, results may not generalizable to other populaitons(2) Responses are self-reported privacy preferences elicited via vignettes or short questionnaires, not observed real-world behaviour, and may exhibit social-desirability bias as well as the well-documented privacy paradox (stated preferences diverging from revealed actions). (3) The data captures privacy attitudes at a single point in time (2025-2026) and does not track norms as they evolve. (4) The dataset is intended as an evaluation benchmark for LLM simulation of user privacy behaviour and is not validated for, and should not be used as, a basis for production privacy decisions, regulatory compliance, or any safety-critical use (clinical, legal, or otherwise).", |
| "rai:dataBiases": "Each contributing study introduces its own population skew, which carries through into the combined dataset:\n- LLM Healthcare Consultation: 846 Chinese adults recruited online; the only non-Western study and the only one without released demographics.\n- AI Agent Permissions: 203 U.S. adults; released demographics cover education, age range, and gender only.\n- LLM Chatbot: 300 U.S. ChatGPT users; demographics not released.\n- Chatbot Group Chat: 374 internationally recruited participants via Prolific (Western-skewed); demographics include age, gender, race/ethnicity, education, and technical-experience flag.\n- LLM Conversational Agents: 277 internationally recruited participants with a heavy European skew; restricted to chatbot users from the original 422-person pool.", |
| "rai:personalSensitiveInformation": "The dataset does not contain sensitive informaiton or any way to link back to the original participants. Even the subset of user studies that release demographics do so in aggregate form (e.g., age ranges rather than individual identifiers).", |
| "rai:dataUseCases": "The dataset is meant to evaluate how well an LLM can simulate a specific human participant's privacy preferences when given that participant's persona facets (demographics, previous AI experiences, stated privacy attitudes) \nOut-of-scope use cases: (1) training LLMs on this dataset — it is intended as evaluation data, with only 1,000-2,000 participants; (2) deploying any system that takes its privacy decisions from the LLM's outputs without human oversight; (2) regulatory compliance assessment, clinical decision support, or any safety-critical privacy decision.", |
| "rai:dataSocialImpact": "Positive impact: PrivacySIM lets researchers and product teams evaluate whether and where LLM-based assistants can faithfully model individual users' privacy preferences, supporting the development of privacy-aware AI systems that anticipate user concerns and reduce the burden of repeated consent or configuration. Public release also improves reproducibility and comparability across LLM evaluations. Negative-impact risks: (1) overestimating LLM capability misread as 'good enough' for production privacy automation; (2) replacing real user input — the dataset must not be used to justify skipping user studies in privacy-sensitive product design; (3) population transfer — privacy norms and language vary across cultures, and inferences should not be extrapolated to populations not represented.", |
| "rai:hasSyntheticData": false, |
| "prov:wasDerivedFrom": [ |
| { |
| "@id": "https://github.com/Cristliu/LLMHealthPrivacy_UserStudy", |
| "prov:label": "LLM Healthcare Consultation user study (Liu et al., IEEE S&P 2025)", |
| "sc:license": "https://opensource.org/licenses/MIT" |
| }, |
| { |
| "@id": "https://github.com/llm-platform-security/ai-agent-permissions", |
| "prov:label": "AI Agent Permissions user study (Wu et al., 2025)", |
| "sc:license": "https://creativecommons.org/licenses/by/4.0/" |
| }, |
| { |
| "@id": "https://doi.org/10.7910/DVN/M6ABJ3", |
| "prov:label": "Replication Data for: Understanding Privacy Norms Around LLM-Based Chatbots (Tran et al., AIES 2025)", |
| "sc:license": "https://creativecommons.org/publicdomain/zero/1.0/" |
| }, |
| { |
| "@id": "https://github.com/csienslab/bot-among-us", |
| "prov:label": "Chatbot Group Chat user study (Chou et al., PoPETs 2026)", |
| "sc:license": "https://opensource.org/licenses/MIT" |
| }, |
| { |
| "@id": "https://osf.io/2vqws/", |
| "prov:label": "LLM Conversational Agents user study (Zufferey et al., PoPETs 2025)", |
| "sc:license": "https://creativecommons.org/licenses/by/4.0/" |
| } |
| ], |
| "prov:wasGeneratedBy": [ |
| { |
| "@type": "prov:Activity", |
| "prov:type": { |
| "@id": "https://www.wikidata.org/wiki/Q4929239" |
| }, |
| "prov:label": "Original user-study data collection", |
| "sc:description": "Each of the five upstream studies independently recruited participants and elicited responses to its own privacy-focused questionnaire. PrivacySIM does not collect any new participant data; all responses are inherited verbatim from the upstream studies. See THIRD_PARTY_LICENSES.md and the upstream publications for per-study recruitment, ethics-review, and compensation details." |
| }, |
| { |
| "@type": "prov:Activity", |
| "prov:type": { |
| "@id": "https://www.wikidata.org/wiki/Q5227332" |
| }, |
| "prov:label": "Schema unification and prompt expansion", |
| "sc:description": "Per-study CSVs are reshaped into a single unified schema (user_id, domain, demographics, previous_experiences, privacy_attitudes, questionnaire, responses, preferences). A deterministic uuid5(NAMESPACE_URL, '<user_id>||<domain>') identifier is added." |
| } |
| ] |
| } |
|
|
