---
license: cc-by-nc-4.0
task_categories:
  - tabular-classification
  - text-generation
language:
  - en
tags:
  - user-profiles
  - matchmaking
  - demographics
  - anonymized
size_categories:
  - 10K<n<100K
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train.parquet
---

# User Profile Dataset (Anonymized)

## Dataset Description

A collection of **10,748 anonymized user profiles** from a matchmaking/dating platform. Each record contains demographic attributes, lifestyle habits, and preferences. The dataset has been de-identified to satisfy **k-anonymity (k >= 5)**.

> **Note on the dataset viewer**
> For privacy reasons, two attributes — **Age** and **Marital Status** — are intentionally **hidden from the dataset viewer preview**. They are **not deleted** from the dataset itself: a complete 24-column file is provided at `full/train.parquet` and can be downloaded directly (see the *Loading the full dataset* section below).

## Files in this repository

| Path | Columns | Purpose |
|------|---------|---------|
| `data/train.parquet` | 22 | Default config; what the dataset viewer renders and what `load_dataset(...)` returns by default. **Excludes** `Age` and `Marital Status`. |
| `full/train.parquet` | 24 | Complete data including `Age` and `Marital Status`. Not registered as a viewer config — must be downloaded explicitly. |
| `anonymize.py` | — | Script used to generate the anonymized data. |

## Fields

| Field | Type | Description | In viewer |
|-------|------|-------------|-----------|
| Gender | Categorical | male / female | yes |
| Age | Range (5-yr or 10-yr bins) | e.g., "55-59", "60-69" | **hidden** |
| Marital Status | Categorical | Divorce, Widow, Never married, etc. | **hidden** |
| Location | Province/Region | Generalized to province or region level | yes |
| Educational Qualifications | Categorical | Bachelor, Associate, Technical secondary school, etc. | yes |
| Height | Range (5cm or 10cm bins) | e.g., "160-164cm", "170-179cm" | yes |
| Car Purchase Situation | Categorical | Purchased / Not yet purchased / Not disclosed | yes |
| Monthly Salary | Range | e.g., "2000-5000 RMB", "5000-10000 RMB" | yes |
| House Purchase Situation | Categorical | Purchased house / Not disclosed | yes |
| Constellation | Categorical | Zodiac sign | yes |
| Nation | Categorical | Ethnic group | yes |
| Blood Group | Categorical | A/B/O/AB-type | yes |
| Smoking Habits | Categorical | Smoking frequency/attitude | yes |
| Drinking Habits | Categorical | Drinking frequency | yes |
| Exercise Habits | Categorical | Exercise frequency | yes |
| Daily Schedule | Categorical | Sleep/wake routine | yes |
| Housework Level | Categorical | Housework involvement | yes |
| Pet Liking Level | Categorical | Attitude toward pets | yes |
| Industry | Categorical | Employment sector | yes |
| Company Type | Categorical | Type of employer | yes |
| Hometown | Province/Region | Generalized to province or region level | yes |
| About Children | Categorical | Preference regarding children | yes |
| Living With Parents | Categorical | Attitude toward living with parents | yes |
| Ranking At Home | Categorical | Birth order | yes |

## Privacy & Anonymization

This dataset has been de-identified using the following techniques:

1. **ID Removal**: Original user IDs have been removed and records shuffled randomly.
2. **Quasi-identifier Generalization** (multi-pass):
   - Age: exact age -> 5-year bins (10-year for harder-to-anonymize records)
   - Height: exact cm -> 5cm bins (10cm for harder cases)
   - Location/Hometown: city-level -> province-level (region-level for harder cases)
3. **k-Anonymity Guarantee**: Every combination of (Gender, Age, Location, Height) appears in at least 5 records (k >= 5).
4. **Record Suppression**: 321 records (2.9%) that could not satisfy k >= 5 even after maximum generalization were removed.
5. **Terminology Normalization**: "secrecy" values replaced with "Not disclosed".
6. **Viewer Hiding**: `Age` and `Marital Status` are excluded from the default viewer config to reduce surface-level exposure of these sensitive attributes; they remain available in `full/train.parquet`.

## Limitations & Ethical Considerations

- This dataset originates from a real matchmaking platform. Despite anonymization, users should NOT attempt re-identification.
- The demographic distribution skews toward ages 50-65, reflecting the source platform's user base.
- Gender distribution is imbalanced (~83% female).
- This dataset is released for **research purposes only** (e.g., recommendation systems, demographic analysis). Commercial use for profiling individuals is prohibited.
- The anonymization satisfies k-anonymity but NOT l-diversity or t-closeness. Sensitive attributes (salary, marital status) may not be uniformly distributed within equivalence classes.

## Usage

### Default (22 columns, matches the viewer)

```python
from datasets import load_dataset

dataset = load_dataset("tsiiiing/ShiJiJiaYuanDataset")
print(dataset["train"][0])
# Does NOT include `Age` or `Marital Status`.
```

### Loading the full dataset (all 24 columns)

The complete file is stored at `full/train.parquet` and is not part of the default config. Load it explicitly:

```python
from huggingface_hub import hf_hub_download
import pandas as pd

path = hf_hub_download(
    repo_id="tsiiiing/ShiJiJiaYuanDataset",
    filename="full/train.parquet",
    repo_type="dataset",
)
df = pd.read_parquet(path)  # 24 columns, includes Age and Marital Status
```

Or via the `datasets` library by pointing at the file directly:

```python
from datasets import load_dataset

dataset = load_dataset(
    "parquet",
    data_files="hf://datasets/tsiiiing/ShiJiJiaYuanDataset/full/train.parquet",
)
```

Or via the CLI:

```bash
hf download tsiiiing/ShiJiJiaYuanDataset --type dataset --include "full/*"
```

## Citation

If you use this dataset, please cite:

```
@dataset{ShiJiJiaYuanDataset,
  title={What you like, what I am: Online dating recommendation via matching individual preferences with features},
  year={2022},
  note={k-anonymized matchmaking platform user profiles}
}
```