WVS2Persona: Parsed World Values Survey (WVS) Wave 7 records into textual personas

Dataset Description

This dataset contains respondent-level persona descriptions derived from the World Values Survey (WVS) Wave 7 core questionnaire.

Each persona corresponds to one individual survey record. These are not cluster centroids, archetypes, or synthetic group summaries. The persona text is a deterministic natural-language rendering of the respondent's answers to the core WVS questionnaire variables only.

On this repo, the dataset is organized by country as subsets/configs. Each subset contains a single train split.

Only the following columns are released in the Hub version:

• persona_id
• persona

Data Source

The underlying survey source is:

• World Values Survey Wave 7

The persona generation pipeline uses only the core questionnaire sections:

• Social Values, Attitudes & Stereotypes
• Happiness and Well-Being
• Social Capital, Trust & Organizational Membership
• Economic Values
• Corruption
• Migration
• Security
• Postmaterialist Index
• Science & Technology
• Religious Values
• Ethical Values and Norms
• Political Interest & Political Participation
• Political Culture & Political Regimes
• Demographics

No regional modules, contextual variables, or post-core thematic modules are used for the released personas.

Methodology

The methodology followed to create these personas is aligned with the approach described in the following paper:

NileChat: Towards Linguistically Diverse and Culturally Aware LLMs for Local Communities

• ACL Anthology: https://aclanthology.org/2025.emnlp-main.556/

In this dataset:

1. Each WVS respondent remains a separate record.
2. Only core questionnaire variables are used.
3. Structured survey values are decoded into human-readable labels.
4. A deterministic persona description is generated from the decoded responses.

A concise / summarized persona version, following the NileChat-style approach, will be provided in a future commit.

Dataset Structure

Fields

• persona_id: Stable identifier for the persona record.
• persona: A full English persona description grounded in the respondent's WVS Wave 7 core-variable responses.

Example

{
    "persona_id": "MAR_504720001",
    "persona": "This person lives in Morocco. She is female, was born in 1969, and is 52 years old. ..."
}

Loading the Dataset

from datasets import load_dataset

REPO_ID = "3ebdola/wvs2persona"

ds_morocco = load_dataset(REPO_ID, "Morocco", split="train")
print(ds_morocco)
print(ds_morocco[0])

Another example:

from datasets import load_dataset

REPO_ID = "3ebdola/wvs2persona"

ds_egypt = load_dataset(REPO_ID, "Egypt", split="train")
print(ds_egypt[0]["persona_id"])
print(ds_egypt[0]["persona"])

Subsets

The dataset currently includes 66 country subsets and 97,220 personas in total.

Subset list and row counts

Subset	Rows
Andorra	1004
Argentina	1003
Armenia	1223
Australia	1813
Bangladesh	1200
Bolivia	2067
Brazil	1762
Canada	4018
Chile	1000
China	3036
Colombia	1520
Cyprus	1000
Czechia	1200
Ecuador	1200
Egypt	1200
Ethiopia	1230
Germany	1528
Great Britain	2609
Greece	1200
Guatemala	1229
Hong Kong SAR	2075
India	1692
Indonesia	3200
Iran	1499
Iraq	1200
Japan	1353
Jordan	1203
Kazakhstan	1276
Kenya	1266
Kyrgyzstan	1200
Lebanon	1200
Libya	1196
Macau SAR	1023
Malaysia	1313
Maldives	1039
Mexico	1741
Mongolia	1638
Morocco	1200
Myanmar	1200
Netherlands	2145
New Zealand	1057
Nicaragua	1200
Nigeria	1237
Northern Ireland	447
Pakistan	1995
Peru	1400
Philippines	1200
Puerto Rico	1127
Romania	1257
Russia	1810
Serbia	1046
Singapore	2012
Slovakia	1200
South Korea	1245
Taiwan ROC	1223
Tajikistan	1200
Thailand	1500
Tunisia	1208
Turkey	2415
Ukraine	1289
United States	2596
Uruguay	1000
Uzbekistan	1250
Venezuela	1190
Vietnam	1200
Zimbabwe	1215

Intended Use

This dataset can be useful for:

• persona-based prompting and conditioning
• culture-aware or country-aware LLM experimentation
• evaluation of value-sensitive or socially grounded generation
• retrieval and few-shot selection by country-specific persona text
• downstream summarization or compression of long persona descriptions

Limitations

• The personas are generated textual summaries, not verbatim respondent statements.
• They are grounded in survey answers, but should not be treated as complete biographies.
• The released text is in English, even when the original respondents come from non-English-speaking countries.
• Persona descriptions may reflect survey instrument limitations, response noise, or country-specific coding artifacts.
• Only the core WVS variables are used in this release.

Citation

If you use this dataset, please cite the following paper:

@inproceedings{el-mekki-etal-2025-nilechat,
    title = "{N}ile{C}hat: Towards Linguistically Diverse and Culturally Aware {LLM}s for Local Communities",
    author = "El Mekki, Abdellah  and
      Atou, Houdaifa  and
      Nacar, Omer  and
      Shehata, Shady  and
      Abdul-Mageed, Muhammad",
    editor = "Christodoulopoulos, Christos  and
      Chakraborty, Tanmoy  and
      Rose, Carolyn  and
      Peng, Violet",
    booktitle = "Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing",
    month = nov,
    year = "2025",
    address = "Suzhou, China",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2025.emnlp-main.556/",
    doi = "10.18653/v1/2025.emnlp-main.556",
    pages = "10967--10991",
    ISBN = "979-8-89176-332-6"
}