---
language:
- en
pretty_name: Melbourne Cafés & Restaurants (OSM / Overpass)
task_categories:
- text-classification
- feature-extraction
tags:
- openstreetmap
- geospatial
- hospitality
- melbourne
- australia
- poi
size_categories:
- 1K-10K
license: odbl
---

---
license: odbl-1.0
---

# Melbourne Cafés & Restaurants (OpenStreetMap / Overpass)

**Area:** Melbourne CBD + Fitzroy + Carlton + Collingwood + Richmond West  
**Method:** Two Overpass API queries (`amenity=cafe`, `amenity=restaurant`) over a Melbourne **bounding box**, merged and augmented  
**Collection date:** 22 October 2025 (Australia/Perth)  
**Canonical file:** `melbourne_cafes_restaurants_clean.csv` (1,896 × 31)  
**Licence:** © OpenStreetMap contributors — **ODbL 1.0**

---

---

A curated dataset of 1,896 cafés and restaurants in Melbourne, extracted from OpenStreetMap via the Overpass API and enriched with engineered features for geospatial analysis, zero‑shot classification, and embedding‑based similarity search.

## Dataset Description
An open dataset of cafés and restaurants in metropolitan Melbourne collected with the OpenStreetMap Overpass API.  
The table includes geospatial coordinates, address details, and engineered features for text and vector-search tasks.

---

---

## Intended Uses

- **Task 1:** Exploratory and geospatial analysis of Melbourne’s café and restaurant landscape.
- **Task 2:** Zero-shot labeling with pre-trained transformers (e.g., “fine dining”, “family friendly”).
- **Task 3:** Text embedding generation and FAISS top‑k similarity search for recommendation or retrieval applications.

---

---

## Collection Parameters
- **Bounding box:** `south = -38.26`, `north = -37.47`, `west = 144.49`, `east = 145.49`
- **Queries executed:**  
  - `amenity="cafe"`  
  - `amenity="restaurant"`
- **Elements captured:** nodes, ways, relations (with `center` geometry)  
- **Output:** JSON → flattened to CSV

**Example Overpass query (cafés):**
```overpass
[out:json][timeout:180];
(
  node["amenity"="cafe"](-38.26,144.49,-37.47,145.49);
  way["amenity"="cafe"](-38.26,144.49,-37.47,145.49);
  relation["amenity"="cafe"](-38.26,144.49,-37.47,145.49);
);
out tags center;
```

---

---

## Data Cleaning & Augmentation

After merging café and restaurant records:

Cleaning: drop null names/coords; de-duplicate by name + rounded coordinates; standardise address fields and combine into address_full.

Derived features:

distance_to_cbd_km — Haversine distance from Flinders Street Station

cuisine_primary — first value from semicolon-separated cuisine

name_len — character length of venue name

desc — short human-readable summary

text — concatenated string used for zero-shot and embeddings

boolean flags: has_wifi, has_outdoor, is_accessible (where inferable)

Validation: simple checks for duplicates, coordinate ranges, and missing essentials.

---

---

## Schema

| Field                | Type      | Description |
|----------------------|-----------|-------------|
| name                 | string    | Business name |
| lat, lon             | float     | Coordinates |
| address              | string    | Street address |
| cuisine_primary      | string    | Extracted primary cuisine type |
| has_wifi             | boolean   | Wi-Fi availability |
| is_accessible        | boolean   | Wheelchair accessibility |
| outdoor_seating      | boolean   | Outdoor seating availability |
| distance_to_cbd_km   | float     | Haversine distance to Melbourne CBD |
| desc                 | string    | Human-readable summary |
| text                 | string    | Concatenated text used for zero-shot and embeddings |

---

---

## Example Use Cases
- Recommendation systems
- Accessibility-aware search
- RAG chatbot
- Market analysis

---

---

## Reliability & Coverage

Coverage bias: OSM completeness varies by suburb; some tags (e.g., cuisine, opening_hours) may be missing.

Temporal snapshot: Data reflects 22 Oct 2025; venues may change over time.

No personal data: Only business POIs (public venues).

Intended Use

Task 2: Apply a pre-trained Transformer (zero-shot) to label text (e.g., fine dining, family friendly).

Task 3: Generate text embeddings and perform FAISS top-k similarity search.

---

---

## Licence & Attribution

ODbL 1.0 — © OpenStreetMap contributors.
Attribution and share-alike apply to adapted databases and redistributions.
See: https://www.openstreetmap.org/copyright

---

---

## Ethics & Privacy (Australia)

Dataset contains only business locations — no personal or sensitive information.
Complies with APP 3 (collection), APP 5 (notification), and APP 8 (cross-border disclosure) under the Privacy Act 1988 (Cth).
Hugging Face hosting occurs outside Australia; attribution, data retention, and provenance are documented in the course report.

---

---

## Storage & Versioning

Earlier raw/intermediate files (osm_melb_cafes_raw.json, melbourne_cafes_restaurants.csv) were uploaded temporarily for validation and are now archived locally to keep the preview clean.
Only the canonical file melbourne_cafes_restaurants_clean.csv remains at repo root.
Version history is available in the Files & Versions tab.

Load
from datasets import load_dataset
ds = load_dataset("CITTYJAMES/melbourne-cafes-restaurants-osm")

---

---

## Citation

Gaddi, Heidi (2025). Melbourne Cafés & Restaurants (OpenStreetMap / Overpass).
Hugging Face Datasets. https://huggingface.co/datasets/CITTYJAMES/melbourne-cafes-restaurants-osm

---

---

## Usage
```python
from datasets import load_dataset
ds = load_dataset("CITTYJAMES/melbourne-cafes-restaurants-osm")