Regenerate croissant.json from de-anonymized repo (parquet file objects, record sets, shortened RAI fields); update dataset card (drop schema notes, arXiv citation)
Browse files- README.md +10 -44
- croissant.json +366 -15
README.md
CHANGED
|
@@ -33,35 +33,6 @@ LLM chatbots, LLM healthcare consultations, AI agent permissions, group chat
|
|
| 33 |
with bots, and conversational agents. Identifiers are derived
|
| 34 |
deterministically from `(user_id, domain)` via UUID5.
|
| 35 |
|
| 36 |
-
## Files
|
| 37 |
-
|
| 38 |
-
| File | Rows | Purpose |
|
| 39 |
-
|---|---|---|
|
| 40 |
-
| `full_dataset.csv` | 2,000 | All users from all five studies, no filtering or sampling. |
|
| 41 |
-
| `dataset.csv` | 1,000 | Sampled to 200 users per domain (healthcare filtered to records with informative previous-experience values), `random_state=42`. |
|
| 42 |
-
| `croissant.json` | | Croissant 1.1 metadata with Responsible AI (RAI) fields describing both CSV files. |
|
| 43 |
-
| `THIRD_PARTY_LICENSES.md` | | Attribution, license texts, and modification notices for the five source studies. |
|
| 44 |
-
|
| 45 |
-
## Schema notes
|
| 46 |
-
|
| 47 |
-
Several columns (`demographics`, `previous_experiences`, `privacy_attitudes`,
|
| 48 |
-
`responses`, `preferences`) hold dict-shaped values. Four of the five
|
| 49 |
-
contributing studies stored these as Python dict reprs (single-quoted);
|
| 50 |
-
the fifth uses JSON. Both forms parse with `ast.literal_eval`:
|
| 51 |
-
|
| 52 |
-
```python
|
| 53 |
-
import ast
|
| 54 |
-
import pandas as pd
|
| 55 |
-
|
| 56 |
-
df = pd.read_csv("dataset.csv")
|
| 57 |
-
prefs = ast.literal_eval(df.iloc[0]["preferences"])
|
| 58 |
-
# -> {question_text: user_response, ...}
|
| 59 |
-
```
|
| 60 |
-
|
| 61 |
-
A column-by-column schema for every file is documented at
|
| 62 |
-
[docs/SCHEMA.md](https://github.com/james-flemings/PrivacySIM/blob/main/docs/SCHEMA.md)
|
| 63 |
-
in the accompanying code repository.
|
| 64 |
-
|
| 65 |
## Loading
|
| 66 |
|
| 67 |
Each configuration exposes a single `train` split containing every row of the
|
|
@@ -78,15 +49,6 @@ ds = load_dataset("jamesflemings/PrivacySIM", "full_dataset", split="train")
|
|
| 78 |
ds = load_dataset("jamesflemings/PrivacySIM", "sampled", split="train")
|
| 79 |
```
|
| 80 |
|
| 81 |
-
## Croissant metadata
|
| 82 |
-
|
| 83 |
-
A [Croissant](https://mlcommons.org/croissant/) 1.1 JSON-LD file that
|
| 84 |
-
describes both CSV files, their fields, the license, and the Responsible AI
|
| 85 |
-
(RAI) and provenance metadata is shipped as
|
| 86 |
-
[`croissant.json`](./croissant.json). The Hub also serves auto-generated
|
| 87 |
-
Croissant metadata at
|
| 88 |
-
`https://huggingface.co/api/datasets/jamesflemings/PrivacySIM/croissant`.
|
| 89 |
-
|
| 90 |
## Source studies
|
| 91 |
|
| 92 |
The data is derived from five public user studies. Each row's `domain`
|
|
@@ -131,13 +93,17 @@ that attribution file (or an equivalent).
|
|
| 131 |
|
| 132 |
## Citation
|
| 133 |
|
| 134 |
-
If you use PrivacySIM, please cite the paper
|
|
|
|
| 135 |
|
| 136 |
```bibtex
|
| 137 |
-
@
|
| 138 |
-
title
|
| 139 |
-
author
|
| 140 |
-
|
| 141 |
-
year
|
|
|
|
|
|
|
|
|
|
| 142 |
}
|
| 143 |
```
|
|
|
|
| 33 |
with bots, and conversational agents. Identifiers are derived
|
| 34 |
deterministically from `(user_id, domain)` via UUID5.
|
| 35 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 36 |
## Loading
|
| 37 |
|
| 38 |
Each configuration exposes a single `train` split containing every row of the
|
|
|
|
| 49 |
ds = load_dataset("jamesflemings/PrivacySIM", "sampled", split="train")
|
| 50 |
```
|
| 51 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 52 |
## Source studies
|
| 53 |
|
| 54 |
The data is derived from five public user studies. Each row's `domain`
|
|
|
|
| 93 |
|
| 94 |
## Citation
|
| 95 |
|
| 96 |
+
If you use PrivacySIM, please cite the paper
|
| 97 |
+
([arXiv:2605.12147](https://arxiv.org/abs/2605.12147)) and the source studies above:
|
| 98 |
|
| 99 |
```bibtex
|
| 100 |
+
@article{flemings2026privacysim,
|
| 101 |
+
title = {{PrivacySIM}: Evaluating {LLM} Simulation of User Privacy Behavior},
|
| 102 |
+
author = {Flemings, James and Annavaram, Murali},
|
| 103 |
+
journal = {arXiv preprint arXiv:2605.12147},
|
| 104 |
+
year = {2026},
|
| 105 |
+
eprint = {2605.12147},
|
| 106 |
+
archivePrefix = {arXiv},
|
| 107 |
+
primaryClass = {cs.CR}
|
| 108 |
}
|
| 109 |
```
|
croissant.json
CHANGED
|
@@ -54,22 +54,375 @@
|
|
| 54 |
"@id": "repo",
|
| 55 |
"name": "repo",
|
| 56 |
"description": "The Hugging Face git repository.",
|
| 57 |
-
"contentUrl": "https://huggingface.co/datasets/
|
| 58 |
"encodingFormat": "git+https",
|
| 59 |
"sha256": "https://github.com/mlcommons/croissant/issues/80"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 60 |
}
|
| 61 |
],
|
| 62 |
-
"recordSet": [],
|
| 63 |
"conformsTo": "http://mlcommons.org/croissant/1.1",
|
| 64 |
"name": "PrivacySIM",
|
| 65 |
-
"description": "\n\t\n\t\t\n\t\
|
| 66 |
"alternateName": [
|
| 67 |
-
"
|
| 68 |
],
|
| 69 |
"creator": {
|
| 70 |
"@type": "Person",
|
| 71 |
-
"name": "
|
| 72 |
-
"url": "https://huggingface.co/
|
| 73 |
},
|
| 74 |
"keywords": [
|
| 75 |
"English",
|
|
@@ -78,18 +431,16 @@
|
|
| 78 |
"🇺🇸 Region: US",
|
| 79 |
"privacy",
|
| 80 |
"llm-evaluation",
|
| 81 |
-
"
|
| 82 |
-
"privacy-norms",
|
| 83 |
-
"user-study",
|
| 84 |
"alignment"
|
| 85 |
],
|
| 86 |
"license": "https://choosealicense.com/licenses/cc-by-4.0/",
|
| 87 |
-
"url": "https://huggingface.co/datasets/
|
| 88 |
-
"rai:dataLimitations": "
|
| 89 |
-
"rai:dataBiases": "
|
| 90 |
-
"rai:personalSensitiveInformation": "The dataset
|
| 91 |
-
"rai:dataUseCases": "
|
| 92 |
-
"rai:dataSocialImpact": "Positive impact:
|
| 93 |
"rai:hasSyntheticData": false,
|
| 94 |
"prov:wasDerivedFrom": [
|
| 95 |
{
|
|
|
|
| 54 |
"@id": "repo",
|
| 55 |
"name": "repo",
|
| 56 |
"description": "The Hugging Face git repository.",
|
| 57 |
+
"contentUrl": "https://huggingface.co/datasets/jamesflemings/PrivacySIM/tree/refs%2Fconvert%2Fparquet",
|
| 58 |
"encodingFormat": "git+https",
|
| 59 |
"sha256": "https://github.com/mlcommons/croissant/issues/80"
|
| 60 |
+
},
|
| 61 |
+
{
|
| 62 |
+
"@type": "cr:FileSet",
|
| 63 |
+
"@id": "parquet-files-for-config-full_dataset",
|
| 64 |
+
"containedIn": {
|
| 65 |
+
"@id": "repo"
|
| 66 |
+
},
|
| 67 |
+
"encodingFormat": "application/x-parquet",
|
| 68 |
+
"includes": "full_dataset/*/*.parquet"
|
| 69 |
+
},
|
| 70 |
+
{
|
| 71 |
+
"@type": "cr:FileSet",
|
| 72 |
+
"@id": "parquet-files-for-config-sampled",
|
| 73 |
+
"containedIn": {
|
| 74 |
+
"@id": "repo"
|
| 75 |
+
},
|
| 76 |
+
"encodingFormat": "application/x-parquet",
|
| 77 |
+
"includes": "sampled/*/*.parquet"
|
| 78 |
+
}
|
| 79 |
+
],
|
| 80 |
+
"recordSet": [
|
| 81 |
+
{
|
| 82 |
+
"@type": "cr:RecordSet",
|
| 83 |
+
"dataType": "cr:Split",
|
| 84 |
+
"key": {
|
| 85 |
+
"@id": "full_dataset_splits/split_name"
|
| 86 |
+
},
|
| 87 |
+
"@id": "full_dataset_splits",
|
| 88 |
+
"name": "full_dataset_splits",
|
| 89 |
+
"description": "Splits for the full_dataset config.",
|
| 90 |
+
"field": [
|
| 91 |
+
{
|
| 92 |
+
"@type": "cr:Field",
|
| 93 |
+
"@id": "full_dataset_splits/split_name",
|
| 94 |
+
"dataType": "sc:Text"
|
| 95 |
+
}
|
| 96 |
+
],
|
| 97 |
+
"data": [
|
| 98 |
+
{
|
| 99 |
+
"full_dataset_splits/split_name": "train"
|
| 100 |
+
}
|
| 101 |
+
]
|
| 102 |
+
},
|
| 103 |
+
{
|
| 104 |
+
"@type": "cr:RecordSet",
|
| 105 |
+
"@id": "full_dataset",
|
| 106 |
+
"description": "jamesflemings/PrivacySIM - 'full_dataset' subset",
|
| 107 |
+
"field": [
|
| 108 |
+
{
|
| 109 |
+
"@type": "cr:Field",
|
| 110 |
+
"@id": "full_dataset/split",
|
| 111 |
+
"dataType": "sc:Text",
|
| 112 |
+
"source": {
|
| 113 |
+
"fileSet": {
|
| 114 |
+
"@id": "parquet-files-for-config-full_dataset"
|
| 115 |
+
},
|
| 116 |
+
"extract": {
|
| 117 |
+
"fileProperty": "fullpath"
|
| 118 |
+
},
|
| 119 |
+
"transform": {
|
| 120 |
+
"regex": "full_dataset/(?:partial-)?(train)/.+parquet$"
|
| 121 |
+
}
|
| 122 |
+
},
|
| 123 |
+
"references": {
|
| 124 |
+
"field": {
|
| 125 |
+
"@id": "full_dataset_splits/split_name"
|
| 126 |
+
}
|
| 127 |
+
}
|
| 128 |
+
},
|
| 129 |
+
{
|
| 130 |
+
"@type": "cr:Field",
|
| 131 |
+
"@id": "full_dataset/uuid",
|
| 132 |
+
"dataType": "sc:Text",
|
| 133 |
+
"source": {
|
| 134 |
+
"fileSet": {
|
| 135 |
+
"@id": "parquet-files-for-config-full_dataset"
|
| 136 |
+
},
|
| 137 |
+
"extract": {
|
| 138 |
+
"column": "uuid"
|
| 139 |
+
}
|
| 140 |
+
}
|
| 141 |
+
},
|
| 142 |
+
{
|
| 143 |
+
"@type": "cr:Field",
|
| 144 |
+
"@id": "full_dataset/user_id",
|
| 145 |
+
"dataType": "cr:Int64",
|
| 146 |
+
"source": {
|
| 147 |
+
"fileSet": {
|
| 148 |
+
"@id": "parquet-files-for-config-full_dataset"
|
| 149 |
+
},
|
| 150 |
+
"extract": {
|
| 151 |
+
"column": "user_id"
|
| 152 |
+
}
|
| 153 |
+
}
|
| 154 |
+
},
|
| 155 |
+
{
|
| 156 |
+
"@type": "cr:Field",
|
| 157 |
+
"@id": "full_dataset/domain",
|
| 158 |
+
"dataType": "sc:Text",
|
| 159 |
+
"source": {
|
| 160 |
+
"fileSet": {
|
| 161 |
+
"@id": "parquet-files-for-config-full_dataset"
|
| 162 |
+
},
|
| 163 |
+
"extract": {
|
| 164 |
+
"column": "domain"
|
| 165 |
+
}
|
| 166 |
+
}
|
| 167 |
+
},
|
| 168 |
+
{
|
| 169 |
+
"@type": "cr:Field",
|
| 170 |
+
"@id": "full_dataset/demographics",
|
| 171 |
+
"dataType": "sc:Text",
|
| 172 |
+
"source": {
|
| 173 |
+
"fileSet": {
|
| 174 |
+
"@id": "parquet-files-for-config-full_dataset"
|
| 175 |
+
},
|
| 176 |
+
"extract": {
|
| 177 |
+
"column": "demographics"
|
| 178 |
+
}
|
| 179 |
+
}
|
| 180 |
+
},
|
| 181 |
+
{
|
| 182 |
+
"@type": "cr:Field",
|
| 183 |
+
"@id": "full_dataset/previous_experiences",
|
| 184 |
+
"dataType": "sc:Text",
|
| 185 |
+
"source": {
|
| 186 |
+
"fileSet": {
|
| 187 |
+
"@id": "parquet-files-for-config-full_dataset"
|
| 188 |
+
},
|
| 189 |
+
"extract": {
|
| 190 |
+
"column": "previous_experiences"
|
| 191 |
+
}
|
| 192 |
+
}
|
| 193 |
+
},
|
| 194 |
+
{
|
| 195 |
+
"@type": "cr:Field",
|
| 196 |
+
"@id": "full_dataset/privacy_attitudes",
|
| 197 |
+
"dataType": "sc:Text",
|
| 198 |
+
"source": {
|
| 199 |
+
"fileSet": {
|
| 200 |
+
"@id": "parquet-files-for-config-full_dataset"
|
| 201 |
+
},
|
| 202 |
+
"extract": {
|
| 203 |
+
"column": "privacy_attitudes"
|
| 204 |
+
}
|
| 205 |
+
}
|
| 206 |
+
},
|
| 207 |
+
{
|
| 208 |
+
"@type": "cr:Field",
|
| 209 |
+
"@id": "full_dataset/questionnaire",
|
| 210 |
+
"dataType": "sc:Text",
|
| 211 |
+
"source": {
|
| 212 |
+
"fileSet": {
|
| 213 |
+
"@id": "parquet-files-for-config-full_dataset"
|
| 214 |
+
},
|
| 215 |
+
"extract": {
|
| 216 |
+
"column": "questionnaire"
|
| 217 |
+
}
|
| 218 |
+
}
|
| 219 |
+
},
|
| 220 |
+
{
|
| 221 |
+
"@type": "cr:Field",
|
| 222 |
+
"@id": "full_dataset/responses",
|
| 223 |
+
"dataType": "sc:Text",
|
| 224 |
+
"source": {
|
| 225 |
+
"fileSet": {
|
| 226 |
+
"@id": "parquet-files-for-config-full_dataset"
|
| 227 |
+
},
|
| 228 |
+
"extract": {
|
| 229 |
+
"column": "responses"
|
| 230 |
+
}
|
| 231 |
+
}
|
| 232 |
+
},
|
| 233 |
+
{
|
| 234 |
+
"@type": "cr:Field",
|
| 235 |
+
"@id": "full_dataset/preferences",
|
| 236 |
+
"dataType": "sc:Text",
|
| 237 |
+
"source": {
|
| 238 |
+
"fileSet": {
|
| 239 |
+
"@id": "parquet-files-for-config-full_dataset"
|
| 240 |
+
},
|
| 241 |
+
"extract": {
|
| 242 |
+
"column": "preferences"
|
| 243 |
+
}
|
| 244 |
+
}
|
| 245 |
+
}
|
| 246 |
+
]
|
| 247 |
+
},
|
| 248 |
+
{
|
| 249 |
+
"@type": "cr:RecordSet",
|
| 250 |
+
"dataType": "cr:Split",
|
| 251 |
+
"key": {
|
| 252 |
+
"@id": "sampled_splits/split_name"
|
| 253 |
+
},
|
| 254 |
+
"@id": "sampled_splits",
|
| 255 |
+
"name": "sampled_splits",
|
| 256 |
+
"description": "Splits for the sampled config.",
|
| 257 |
+
"field": [
|
| 258 |
+
{
|
| 259 |
+
"@type": "cr:Field",
|
| 260 |
+
"@id": "sampled_splits/split_name",
|
| 261 |
+
"dataType": "sc:Text"
|
| 262 |
+
}
|
| 263 |
+
],
|
| 264 |
+
"data": [
|
| 265 |
+
{
|
| 266 |
+
"sampled_splits/split_name": "train"
|
| 267 |
+
}
|
| 268 |
+
]
|
| 269 |
+
},
|
| 270 |
+
{
|
| 271 |
+
"@type": "cr:RecordSet",
|
| 272 |
+
"@id": "sampled",
|
| 273 |
+
"description": "jamesflemings/PrivacySIM - 'sampled' subset",
|
| 274 |
+
"field": [
|
| 275 |
+
{
|
| 276 |
+
"@type": "cr:Field",
|
| 277 |
+
"@id": "sampled/split",
|
| 278 |
+
"dataType": "sc:Text",
|
| 279 |
+
"source": {
|
| 280 |
+
"fileSet": {
|
| 281 |
+
"@id": "parquet-files-for-config-sampled"
|
| 282 |
+
},
|
| 283 |
+
"extract": {
|
| 284 |
+
"fileProperty": "fullpath"
|
| 285 |
+
},
|
| 286 |
+
"transform": {
|
| 287 |
+
"regex": "sampled/(?:partial-)?(train)/.+parquet$"
|
| 288 |
+
}
|
| 289 |
+
},
|
| 290 |
+
"references": {
|
| 291 |
+
"field": {
|
| 292 |
+
"@id": "sampled_splits/split_name"
|
| 293 |
+
}
|
| 294 |
+
}
|
| 295 |
+
},
|
| 296 |
+
{
|
| 297 |
+
"@type": "cr:Field",
|
| 298 |
+
"@id": "sampled/uuid",
|
| 299 |
+
"dataType": "sc:Text",
|
| 300 |
+
"source": {
|
| 301 |
+
"fileSet": {
|
| 302 |
+
"@id": "parquet-files-for-config-sampled"
|
| 303 |
+
},
|
| 304 |
+
"extract": {
|
| 305 |
+
"column": "uuid"
|
| 306 |
+
}
|
| 307 |
+
}
|
| 308 |
+
},
|
| 309 |
+
{
|
| 310 |
+
"@type": "cr:Field",
|
| 311 |
+
"@id": "sampled/user_id",
|
| 312 |
+
"dataType": "cr:Int64",
|
| 313 |
+
"source": {
|
| 314 |
+
"fileSet": {
|
| 315 |
+
"@id": "parquet-files-for-config-sampled"
|
| 316 |
+
},
|
| 317 |
+
"extract": {
|
| 318 |
+
"column": "user_id"
|
| 319 |
+
}
|
| 320 |
+
}
|
| 321 |
+
},
|
| 322 |
+
{
|
| 323 |
+
"@type": "cr:Field",
|
| 324 |
+
"@id": "sampled/domain",
|
| 325 |
+
"dataType": "sc:Text",
|
| 326 |
+
"source": {
|
| 327 |
+
"fileSet": {
|
| 328 |
+
"@id": "parquet-files-for-config-sampled"
|
| 329 |
+
},
|
| 330 |
+
"extract": {
|
| 331 |
+
"column": "domain"
|
| 332 |
+
}
|
| 333 |
+
}
|
| 334 |
+
},
|
| 335 |
+
{
|
| 336 |
+
"@type": "cr:Field",
|
| 337 |
+
"@id": "sampled/demographics",
|
| 338 |
+
"dataType": "sc:Text",
|
| 339 |
+
"source": {
|
| 340 |
+
"fileSet": {
|
| 341 |
+
"@id": "parquet-files-for-config-sampled"
|
| 342 |
+
},
|
| 343 |
+
"extract": {
|
| 344 |
+
"column": "demographics"
|
| 345 |
+
}
|
| 346 |
+
}
|
| 347 |
+
},
|
| 348 |
+
{
|
| 349 |
+
"@type": "cr:Field",
|
| 350 |
+
"@id": "sampled/previous_experiences",
|
| 351 |
+
"dataType": "sc:Text",
|
| 352 |
+
"source": {
|
| 353 |
+
"fileSet": {
|
| 354 |
+
"@id": "parquet-files-for-config-sampled"
|
| 355 |
+
},
|
| 356 |
+
"extract": {
|
| 357 |
+
"column": "previous_experiences"
|
| 358 |
+
}
|
| 359 |
+
}
|
| 360 |
+
},
|
| 361 |
+
{
|
| 362 |
+
"@type": "cr:Field",
|
| 363 |
+
"@id": "sampled/privacy_attitudes",
|
| 364 |
+
"dataType": "sc:Text",
|
| 365 |
+
"source": {
|
| 366 |
+
"fileSet": {
|
| 367 |
+
"@id": "parquet-files-for-config-sampled"
|
| 368 |
+
},
|
| 369 |
+
"extract": {
|
| 370 |
+
"column": "privacy_attitudes"
|
| 371 |
+
}
|
| 372 |
+
}
|
| 373 |
+
},
|
| 374 |
+
{
|
| 375 |
+
"@type": "cr:Field",
|
| 376 |
+
"@id": "sampled/questionnaire",
|
| 377 |
+
"dataType": "sc:Text",
|
| 378 |
+
"source": {
|
| 379 |
+
"fileSet": {
|
| 380 |
+
"@id": "parquet-files-for-config-sampled"
|
| 381 |
+
},
|
| 382 |
+
"extract": {
|
| 383 |
+
"column": "questionnaire"
|
| 384 |
+
}
|
| 385 |
+
}
|
| 386 |
+
},
|
| 387 |
+
{
|
| 388 |
+
"@type": "cr:Field",
|
| 389 |
+
"@id": "sampled/responses",
|
| 390 |
+
"dataType": "sc:Text",
|
| 391 |
+
"source": {
|
| 392 |
+
"fileSet": {
|
| 393 |
+
"@id": "parquet-files-for-config-sampled"
|
| 394 |
+
},
|
| 395 |
+
"extract": {
|
| 396 |
+
"column": "responses"
|
| 397 |
+
}
|
| 398 |
+
}
|
| 399 |
+
},
|
| 400 |
+
{
|
| 401 |
+
"@type": "cr:Field",
|
| 402 |
+
"@id": "sampled/preferences",
|
| 403 |
+
"dataType": "sc:Text",
|
| 404 |
+
"source": {
|
| 405 |
+
"fileSet": {
|
| 406 |
+
"@id": "parquet-files-for-config-sampled"
|
| 407 |
+
},
|
| 408 |
+
"extract": {
|
| 409 |
+
"column": "preferences"
|
| 410 |
+
}
|
| 411 |
+
}
|
| 412 |
+
}
|
| 413 |
+
]
|
| 414 |
}
|
| 415 |
],
|
|
|
|
| 416 |
"conformsTo": "http://mlcommons.org/croissant/1.1",
|
| 417 |
"name": "PrivacySIM",
|
| 418 |
+
"description": "\n\t\n\t\t\n\t\n\t\n\t\tPrivacySIM: An Evaluation Suite for LLM Simulation of User Privacy Behavior\n\t\n\nPrivacySIM aggregates user privacy preferences from five published user\nstudies into a unified evaluation suite for LLM simulation of user\nprivacy behavior. Each row represents one participant's responses to a\nstudy-specific questionnaire about whether data-sharing involving an LLM\nis appropriate, plus persona facets (demographics, previous experience, privacy attitudes).\nThe five contributing studies… See the full description on the dataset page: https://huggingface.co/datasets/jamesflemings/PrivacySIM.",
|
| 419 |
"alternateName": [
|
| 420 |
+
"jamesflemings/PrivacySIM"
|
| 421 |
],
|
| 422 |
"creator": {
|
| 423 |
"@type": "Person",
|
| 424 |
+
"name": "James Flemings",
|
| 425 |
+
"url": "https://huggingface.co/jamesflemings"
|
| 426 |
},
|
| 427 |
"keywords": [
|
| 428 |
"English",
|
|
|
|
| 431 |
"🇺🇸 Region: US",
|
| 432 |
"privacy",
|
| 433 |
"llm-evaluation",
|
| 434 |
+
"llm-simulation",
|
|
|
|
|
|
|
| 435 |
"alignment"
|
| 436 |
],
|
| 437 |
"license": "https://choosealicense.com/licenses/cc-by-4.0/",
|
| 438 |
+
"url": "https://huggingface.co/datasets/jamesflemings/PrivacySIM",
|
| 439 |
+
"rai:dataLimitations": "(1) Participants are predominantly English-speaking and skew toward Western, European, and East-Asian populations, so results may not generalize to other populations. (2) Responses are self-reported privacy preferences from questionnaires, not observed behavior, and may reflect the privacy paradox. (3) The data reflects privacy attitudes at a single point in time (2025-2026).",
|
| 440 |
+
"rai:dataBiases": "The combined dataset inherits the population skew of its five source studies. Four studies recruited mostly Western (U.S. and European) participants and one (LLM Healthcare Consultation) recruited Chinese adults, and two of the five studies did not release demographics.",
|
| 441 |
+
"rai:personalSensitiveInformation": "The dataset contains no identifiers that could link rows back to the original participants. Demographics, where released, are coarse (e.g., age ranges).",
|
| 442 |
+
"rai:dataUseCases": "Evaluating how well an LLM can simulate an individual participant's privacy preferences from their persona facets (demographics, previous AI experiences, stated privacy attitudes).",
|
| 443 |
+
"rai:dataSocialImpact": "Positive impact: enables evaluation of whether LLMs can faithfully simulate individual users' privacy preferences, supporting privacy-aware AI systems and reproducible LLM evaluation. Risks: overestimating LLM capability for production privacy automation, using the dataset to justify skipping user studies, and extrapolating privacy norms to populations not represented.",
|
| 444 |
"rai:hasSyntheticData": false,
|
| 445 |
"prov:wasDerivedFrom": [
|
| 446 |
{
|