Algerian-realestate-NER-dataset

Dataset Description

This is a specialized Named Entity Recognition dataset extracted from the complex reality of the Algerian digital real-estate market in Facebook groups, it contains 13 labeled entity and 7138 training example

Real estate advertisements in Algeria (found on Facebook groups) are unstructured , noisy and bloated with code-switching between Algerian Darja (dialect), Arabizi, Standard Arabic, and French, This dataset bridges the gap for low resource dialectal NLP by providing a labeled corpus capable of training state-of-the-art real-estate information extraction models (like AlgerianDeBERTa or dziriBERT) Visit the medium story for more details

• Task: Token Classification / Named Entity Recognition
• Domain: Real Estate
• Languages: Algerian Arabic (Darja), French, Arabizi, Modern Standard Arabic (MSA)

Dataset Overview

The dataset is split into standard Training, Validation, and Test sets

Split	Examples	Purpose
Train	7,138	Primary model training and fine-tuning
Validation	586	Hyperparameter tuning and epoch evaluation
Test	596	Final unseen evaluation benchmark

Entity Schema

The dataset uses a 27 label BIO (Begin, Inside, Outside) schema based on 13 core entity types essential for real-estate information extraction

Entity Type	Description	Examples
PROPERTY_TYPE	Category of the real estate asset	Villa, Appartement, Lot de terrain, Carcasse, Haouch
APT_CLASS	Specific apartment configurations	F3, F4, S+1, Studio
TRANSACTION	The intent or nature of the listing	Vente (Sale), Location (Rent), Echange, خاصني (Looking for)
WILAYA	The province / state (incl. numerical codes)	Alger, Oran, Setif, 16, 31, ولاية
NEIGHBORHOOD	Specific neighborhoods, districts, or streets	Bab Ezzouar, Hai Yasmine, Sidi Yahia
PRICE	Cost or rent amounts, often with local currency/slang	1.5 Milliards, 50000 DA, 200 millions, 800 U
SURFACE	Area dimensions and measurements	120m², 150 metre, 2 hectares
FLOOR	The floor level of the property	1er étage, RDC, 5eme, R+2
CITY	City / commune	Bab Ezzouar · Sidi Yahia · Ain Benian
PHONE	Contact phone numbers (anonymized)	[PHONE]
AMENITY	Features, utilities, or property additions	Garage, Jardin, Ascenseur, Gaz, Bache a eau
DOCUMENT	Legal paperwork and land registry status	Acte notarié, Livret foncier, Papier timbre, تنازل
CONDITION	Physical state or finishing level of the build	Fini, Semi-fini, En construction, a rénover

Privacy and Anonymization

Because the source text originates from Facebook groups, anonymization was applied to protect Personally Identifiable Information:

• Preserved Positional Context: Every token originally tagged as a phone number was replaced with the literal string [PHONE]. The BIO tags (B-PHONE, I-PHONE) remain intact so the model learns the structural positioning of contact info without memorizing real digits

Usage

You can load the dataset using the Hugging Face datasets library:

from datasets import load_dataset, Features, Sequence, ClassLabel, Value


dataset = load_dataset("81melody/algerian-realestate-ner-dataset")

label_list = [f"LABEL_{i}" for i in range(27)]


new_features = Features({
    'tokens': Sequence(Value('string')),
    'ner_tags': Sequence(ClassLabel(names=label_list))
})

dataset = dataset.cast(new_features)

print(f"Features: {dataset['train'].features['ner_tags'].feature.names}")

Data Collection and Annotation

• Source: Data was collected from public Algerian real estate Facebook groups using Facebook API
• Annotation Process: please visit the medium story for more inoformation about the annotation and labeling process
• Challenges: High variance in spelling (Appartement, apartement, برطمة , ابارتومون) and mixed currency formats ( Milliards, U, DA) were standardized under the unified BIO schema

Limitations and Biases

• Platform Bias: The data is sourced exclusively from Facebook, which reflects the casual tone of social media It may not perform as well on highly formal, legal real-estate documents

License

This dataset is licensed under Apache 2.0

Citation

If you use this dataset in your research or applications, please cite it as:

@dataset{algerian_realestate_ner_2026_dataset,
  author       = {Himeur,Ayoub},
  title        = {Algerian-realestate-NER: A dataset for Algerian real-estate information extraction},
  year         = {2026},
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/datasets/81melody/algerian-realestate-ner}
}

Contact

mohamed.himeur@student.unamur.be