Datasets:

ht-totinee
/

JUCampusNews

Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
content_id: int64
main_domain: string
title: string
news_content: string
timestamp: string
category: string
label: int64
content_ID: null
-- schema metadata --
pandas: '{"index_columns": [{"kind": "range", "name": null, "start": 0, "' + 1074
to
{'content_ID': Value('int64'), 'main_domain': Value('string'), 'timestamp': Value('string'), 'category': Value('string'), 'title': Value('string'), 'news_content': Value('string'), 'label': ClassLabel(names=['fake', 'real'])}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 99, in get_rows_or_raise
                  return get_rows(
                         ^^^^^^^^^
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                         ^^^^^^^^^^^^^^^^^^^^^
                File "/src/services/worker/src/worker/utils.py", line 77, in get_rows
                  rows_plus_one = list(itertools.islice(ds, rows_max_number + 1))
                                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2690, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2227, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2260, in _iter_arrow
                  pa_table = cast_table_to_features(pa_table, self.features)
                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2223, in cast_table_to_features
                  raise CastError(
              datasets.table.CastError: Couldn't cast
              content_id: int64
              main_domain: string
              title: string
              news_content: string
              timestamp: string
              category: string
              label: int64
              content_ID: null
              -- schema metadata --
              pandas: '{"index_columns": [{"kind": "range", "name": null, "start": 0, "' + 1074
              to
              {'content_ID': Value('int64'), 'main_domain': Value('string'), 'timestamp': Value('string'), 'category': Value('string'), 'title': Value('string'), 'news_content': Value('string'), 'label': ClassLabel(names=['fake', 'real'])}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

JUCampusNews: A Bangla Campus News Dataset for Disinformation Detection

Dataset Summary

JUCampusNews is a curated Bengali-language dataset of campus news articles collected from e-news portals covering Jahangirnagar University (JU), Bangladesh. It is designed for the task of disinformation (fake news) detection and forensic profiling of news correspondents and their e-news portals.

The dataset contains approximately 1,000 authentic and 100 manually annotated fake news articles in the Bangla (Bengali) language — making it a valuable resource for low-resource NLP research, specifically for Bangla fake news detection.

Dataset Details

Dataset Description

News articles were collected from Bangla-language e-news portals associated with Jahangirnagar University, sourced from the Facebook page "জাবির সকল সংবাদ" (All News of JU). Fake news articles were manually annotated by domain experts. The dataset spans a six-month collection period to address the challenge of news article archival expiry faced by student correspondents.

This dataset was created as part of a Bachelor of Science dissertation at the Department of Computer Science and Engineering, Jahangirnagar University, titled "Disinformation Detection of Campus News and Forensic Profiling of Correspondents and Their E-News Portals".

Curated by: Suraiya Mahmuda & Hasneen Tamanna, Jahangirnagar University
Language: Bangla (Bengali) — bn
License: Creative Commons Attribution 4.0 (CC BY 4.0)

Dataset Structure

Data Instances

Each instance represents a single news article with the following fields:

Field	Type	Description
`content_ID`	int	Unique identifier for each news article
`main_domain`	string	Website or portal that published the news (e.g., `bahannonews.com`)
`timestamp`	string	Publication date and time (e.g., `2025-01-02 11:22:00`)
`category`	string	Thematic category of the news
`title`	string	Headline/title of the article (in Bangla)
`news_content`	string	Full body/content of the article (in Bangla)
`label`	int	`1` = Real/Authentic news, `0` = Fake/Disinformative news

Data Fields — Category Values

The category field contains one of the following values:

Education · Achievement · Crime · Politics · Health · Event · Entertainment · International · Sports · Miscellaneous

Data Splits

The dataset is split 80/20 for training and testing:

Split	Number of Examples
Train	~880
Test	~220

Sample Instance

{
  "content_ID": 1,
  "main_domain": "bahannonews.com",
  "timestamp": "2025-01-02 11:22:00",
  "category": "Education",
  "title": "জাহাঙ্গীরনগর বিশ্ববিদ্যালয়ে ভর্তির আবেদন শুরু আজ",
  "news_content": "জাহাঙ্গীরনগর বিশ্ববিদ্যালয়ের (জাবি) ২০২৪-২৫ শিক্ষাবর্ষের...",
  "label": 1
}

Dataset Creation

Source Data

News articles were collected from multiple Bangla e-news portals covering Jahangirnagar University campus events, spanning January 2025 to July 2025. The primary aggregation source was the Facebook page "জাবির সকল সংবাদ".

Annotation Process

Authentic news (label = 1): ~1,000 articles verified as real news from credible campus news portals.
Fake news (label = 0): ~100 articles manually annotated as disinformative/fake by the research team, based on content verification, source reliability, and cross-referencing with other portals.

Annotation Guidelines

Annotators evaluated articles based on:

Source reliability
Content realism and verifiability
Cross-referencing with other news portals
Existence of misleading or false claims

Uses

Direct Use

This dataset is intended for:

Training and evaluating Bangla fake news / disinformation detection models
Fine-tuning transformer-based models such as BanglaBERT, multilingual BERT, or XLM-RoBERTa on low-resource Bangla NLP tasks
Research on forensic profiling of journalists and news portals based on content authenticity
Benchmarking classical ML classifiers (Logistic Regression, SVM, XGBoost) on Bangla text classification

Out-of-Scope Use

This dataset is campus-specific and may not generalize to national or international Bangla news
Not intended for commercial content moderation without further validation
Should not be used to make automated editorial judgments about individual journalists without human oversight

Dataset Limitations and Biases

Class imbalance: The dataset is heavily skewed — approximately 10:1 ratio of real to fake news. This should be addressed during model training (e.g., SMOTE, class-weighted loss).
Domain specificity: All articles are from a single university campus, limiting generalizability.
Annotation subjectivity: Manual annotation of fake news carries inherent subjectivity.
Archival constraints: Student news portals often unpublish articles within 2–6 months, limiting the volume of collectable data.
Script complexity: Bangla script, grammar, and code-mixing introduce unique NLP challenges not present in high-resource languages.

Experiments and Baseline Results

The following baselines were established by the dataset authors:

Model	Accuracy	Notes
BanglaBERT (fine-tuned)	92%	High accuracy but affected by class imbalance; predicted all as real
LLaMA 2 7B Chat	50%	Predicted all articles as fake; small test set

Note: The high accuracy of BanglaBERT reflects dataset imbalance rather than true detection capability. Future work should incorporate balanced evaluation metrics (macro F1, AUC-ROC).

Citation

If you use this dataset in your research, please cite:

@misc{jucampusnews2025,
  title     = {JUCampusNews: A Bangla Campus News Dataset for Disinformation Detection and Forensic Journalist Profiling},
  author    = {Suraiya Mahmuda, Hasneen Tamanna},
  year      = {2025},
  institution = {Department of Computer Science and Engineering, Jahangirnagar University},
  note      = {Bachelor of Science Dissertation, Faculty of Mathematical and Physical Sciences}
}

Related Work

This dataset was constructed in the context of the following related Bangla NLP and fake news detection works:

Hossain et al. (2021) — BanFakeNews dataset (~50,000 articles), CNN-LSTM hybrid
Rasel et al. (2022) — BanglaBERT, SVM, BiLSTM for Bangla fake news (95%+ accuracy)
Majumdar et al. (2025) — BanFakeNews-2.0 (13,000 articles), BLOOM + m-BERT with QLoRA

Dataset Card Contact

For questions about this dataset, please open a discussion on the Hugging Face dataset repository page.

Downloads last month: 38

Total file size:

5.98 MB