README.md

---
license: cc0-1.0
language:
  - en
pretty_name: MDPI Taxonomy
tags:
  - taxonomy
  - article-classification
  - research-topics
  - openalex
  - hierarchical
  - scholarly
task_categories:
  - text-classification
size_categories:
  - 100K<n<1M
configs:
  - config_name: taxonomy_flat
    data_files: data/taxonomy_flat.parquet
    default: true
  - config_name: taxonomy_hierarchy
    data_files: data/taxonomy_hierarchy.parquet
  - config_name: domain
    data_files: data/domain.parquet
  - config_name: field
    data_files: data/field.parquet
  - config_name: subfield
    data_files: data/subfield.parquet
  - config_name: topic
    data_files: data/topic.parquet
  - config_name: concept
    data_files: data/concept.parquet
---

# MDPI Taxonomy

**Visualization:** [Atlantis](https://atlantis.mdpi.com) is an interactive visualization of this taxonomy with MDPI article classification.

## What is it?

![MDPI Taxonomy hierarchy: Domain → Field → Subfield → Topic → Concept](assets/taxonomy_overview.png)

MDPI Taxonomy is a five-level hierarchical taxonomy for classifying scholarly content across scientific domains. It builds on the [OpenAlex](https://openalex.org/) hierarchy (**Domain → Field → Subfield → Topic**) and extends it with an MDPI **Concept** layer (~113k nodes) for finer article classification and topic tagging.

| Level | Name | Count | Source |
|------:|------|------:|--------|
| 1 | Domain | 4 | OpenAlex |
| 2 | Field | 26 | OpenAlex |
| 3 | Subfield | 252 | OpenAlex |
| 4 | Topic | 4,516 | OpenAlex |
| 5 | Concept | 113,892 | MDPI |

**Domains:** Health Sciences · Life Sciences · Physical Sciences · Social Sciences

Each node has a name and a short explanation. **Explanations are AI-generated** (concise glosses, not expert-curated definitions).

### Worked example

A concept can sit under more than one valid path. Example for **Industry 5.0**:

**Physical Sciences → Engineering → Electrical and Electronic Engineering → Advanced Data and IoT Technologies → Industry 5.0**

![Physical Sciences → Engineering → Electrical and Electronic Engineering → Advanced Data and IoT Technologies → Industry 5.0](assets/taxonomy_hierarchy_example.png)

---

## What is being released

One dataset, multiple configs (one Parquet file each):

| Config | What it is | Best for |
|--------|------------|----------|
| `taxonomy_flat` (**default**) | Denormalized paths: IDs, names, explanations at every level | Browsing and training without joins |
| `taxonomy_hierarchy` | ID links for the full hierarchy | Building trees / graphs |
| `domain` · `field` · `subfield` · `topic` · `concept` | Normalized level tables | Custom joins |

---

## How to download and use it

### 🤗 Datasets

```python
from datasets import load_dataset

ds = load_dataset("mdpi-di/taxonomy", "taxonomy_flat")
print(ds["train"][0])

# Or a single level
concepts = load_dataset("mdpi-di/taxonomy", "concept")
```

### Polars

```python
import polars as pl

flat = pl.read_parquet("data/taxonomy_flat.parquet")

# Recommended: full latest taxonomy (all rows through v5)
latest = flat  # v5 is the latest release; this file is the full snapshot as of v5

health = latest.filter(pl.col("domain_name") == "Health Sciences")
print(health.select(["field_name", "subfield_name", "topic_name", "concept_name"]).head())
```

### Join normalized tables

```python
import polars as pl

domain = pl.read_parquet("data/domain.parquet")
field = pl.read_parquet("data/field.parquet")
subfield = pl.read_parquet("data/subfield.parquet")
topic = pl.read_parquet("data/topic.parquet")
concept = pl.read_parquet("data/concept.parquet")

tree = (
    concept
    .join(topic, on="topic_id", how="left", suffix="_topic")
    .join(subfield, on="subfield_id", how="left", suffix="_subfield")
    .join(field, on="field_id", how="left", suffix="_field")
    .join(domain, on="domain_id", how="left", suffix="_domain")
)
```

---

## Dataset statistics

| Property | Value |
|----------|------:|
| Levels | 5 |
| Concepts | 113,892 |
| Full paths (`taxonomy_flat`) | 113,947 |

### Versions

Rows are tagged with `version` (`v1`, `v3`, `v4`, **`v5`**).

**`v5` is the latest.** We recommend using the **full current taxonomy** in this release (all rows — `v1` through `v5`). Most nodes were added in `v1`; later versions add incremental concepts/paths up to `v5`.

| version | rows | notes |
|---------|-----:|-------|
| v1 | 112,133 | bulk of the taxonomy |
| v3 | 344 | incremental |
| v4 | 501 | incremental |
| **v5** | **969** | **latest increment** |

Filter to a single tag only if you need that increment alone, e.g. `flat.filter(pl.col("version") == "v5")`.

---

## Dataset structure

```text
data/
  domain.parquet
  field.parquet
  subfield.parquet
  topic.parquet
  concept.parquet
  taxonomy_hierarchy.parquet
  taxonomy_flat.parquet
```

| Table | Columns |
|-------|---------|
| `domain` | `domain_id`, `domain_name`, `domain_explanation`, `version`, `created_at` |
| `field` | `field_id`, `domain_id`, `field_name`, `field_explanation`, `version`, `created_at` |
| `subfield` | `subfield_id`, `field_id`, `subfield_name`, `subfield_explanation`, `version`, `created_at` |
| `topic` | `topic_id`, `subfield_id`, `topic_name`, `topic_explanation`, `version`, `created_at` |
| `concept` | `concept_id`, `topic_id`, `concept_name`, `concept_explanation`, `version`, `created_at` |
| `taxonomy_hierarchy` | `id`, `domain_id`, `field_id`, `subfield_id`, `topic_id`, `concept_id`, `version`, `created_at` |
| `taxonomy_flat` | All IDs, names, and explanations for every level, plus `version`, `created_at` |

---

## Dataset creation

- **Levels 1–4** come from the [OpenAlex](https://openalex.org/) topic hierarchy.
- **Level 5 (Concept)** is produced by MDPI Data Intelligence.
- **Explanations** are AI-generated short texts for browsing and labeling — not authoritative definitions.

---

## Considerations for using the data

**Intended uses**
- Article / topic classification and labeling
- Hierarchical retrieval over scientific concepts
- Training or evaluating topic-tagging models

**Limitations**
- AI-generated explanations may be inaccurate
- Multiple `version` values coexist; **prefer the full table as of v5** (all rows). Filter by `version` only for incremental slices
- Some hierarchy rows may have null `concept_id` (path ends at topic)
- A concept may appear under more than one parent path

---

## Additional information

### License

[CC0 1.0 Universal](https://creativecommons.org/publicdomain/zero/1.0/)

### Citation

```bibtex
@dataset{mdpi_taxonomy,
  title        = {MDPI Taxonomy},
  author       = {{MDPI Data Intelligence}},
  year         = {2025},
  note         = {Hierarchical research topic taxonomy (Domain → Field → Subfield → Topic → Concept), extending OpenAlex},
  url          = {https://huggingface.co/datasets/mdpi-di/taxonomy},
  license      = {CC0-1.0}
}
```

### Acknowledgements

Built on the OpenAlex topic hierarchy. Concept-level extensions and packaging by MDPI Data Intelligence.

### Contact

Questions about this dataset or attribution: MDPI Data Intelligence team.