Datasets:

nahid-hub
/

BLUGE-bengali-news-classification

language:
  - bn
pretty_name: 'BLUGE-NCC: Bangla News Classification'
license: cc-by-4.0
task_categories:
  - text-classification
task_ids:
  - topic-classification
tags:
  - bengali
  - bangla
  - bluge
  - ncc
  - news-classification
  - topic-classification
  - bengali-nlp
  - bnlp
  - low-resource
  - text-classification
size_categories:
  - 10K<n<100K
configs:
  - config_name: default
    data_files:
      - split: train
        path: train.parquet
      - split: validation
        path: validation.parquet
      - split: test
        path: test.parquet

BLUGE-NCC: Bangla News Classification

BLUGE-NCC is a meticulously curated and balanced Bangla News Category Classification dataset, one of the 7 tasks in BLUGE (Bengali Language UnderstandinG Evaluation), a balanced benchmark for evaluating Bengali natural language understanding. See the full BLUGE collection for all 7 tasks, and the B-CORE pretraining corpus and BnLM model suite released alongside it.

Dataset Description

This task classifies Bangla news articles into seven categories:

Label	Category
`0`	Business
`1`	Technology
`2`	Crime
`3`	Entertainment
`4`	International Affairs
`5`	Sports
`6`	Lifestyle

Unlike the other BLUGE tasks, NCC was constructed entirely from scratch, using sources independent of the B-CORE pretraining corpus, so there is no train/pretraining data leakage between B-CORE and this evaluation set. The dataset is perfectly class-balanced, with exactly 8,900 samples per category (62,300 total). Labels were derived automatically and consistently by extracting the category directly from each article's source URL, rather than through manual re-labeling — for each URL $u_i$ paired with article $a_i$, the category label $L(u_i)$ was mapped to a normalized label $N(L(u_i))$, yielding article-label pairs.

Dataset Structure

Balanced 80 / 10 / 10 split:

Split	Samples
Train	49,840
Validation	6,230
Test	6,230
Total	62,300

Fields:

text — the Bangla news article content
label — news category (0–6, see table above)

Usage

from datasets import load_dataset

# Load all splits
ds = load_dataset("nahid-hub/BLUGE-bengali-news-classification")

# Access a specific split
train_ds = ds["train"]
val_ds   = ds["validation"]
test_ds  = ds["test"]

print(train_ds[0])

Load a single split directly:

from datasets import load_dataset

test_ds = load_dataset("nahid-hub/BLUGE-bengali-news-classification", split="test")

Or read the Parquet files directly with pandas:

import pandas as pd

train_df = pd.read_parquet(
    "hf://datasets/nahid-hub/BLUGE-bengali-news-classification/train.parquet"
)

Source Data & Attribution

Articles were collected from Bangla news sources independent of B-CORE, with category labels derived automatically from each article's source URL. Please note that while this dataset compilation is released under the license below, the underlying article text may remain subject to the copyright of the original publishing outlets; this release is intended for non-commercial research and benchmarking use consistent with standard fair-use norms for NLP dataset construction. If you are a rights holder with concerns about specific content, please reach out via the repository's community tab.

License

Released under CC BY 4.0 for the dataset compilation, labels, and splits. You are free to share and adapt this dataset for research and most other purposes, provided you give appropriate credit — see the note above regarding underlying article copyright.

Citation

If you use this dataset, please cite:

@ARTICLE{BnLM-BLUGE-B-CORE,
  author={Hossain, Nahid and Faisal Kabir, Md.},
  journal={IEEE Access}, 
  title={Efficient Monolingual Pretraining in Low-Resource Settings Through Morphology-Aware Tokenization, Principled Corpus Denoising, and Benchmark-Driven Evaluation}, 
  year={2026},
  volume={14},
  number={},
  pages={91979-92003},
  keywords={Modeling;Multilingual;Training;Cleaning;Vocabulary;Labeling;Tokenization;Computational linguistics;Pipelines;Conferences;B-CORE;BLUGE;BnLM;corpus;evaluation benchmark;pretrained models},
  doi={10.1109/ACCESS.2026.3701520}
}