Bengali_Crime_Event / README.md
mithunhossain's picture
Update README.md
babd6eb verified
metadata
license: cc-by-4.0
task_categories:
  - text-classification
language:
  - bn
tags:
  - text
  - classification
  - crime
  - event
  - detection
pretty_name: crime
size_categories:
  - 1K<n<10K

About Dataset Bengali Crime Event Dataset – Law & Order, Governance & Social Issues (2025)

This dataset contains Bengali news headlines annotated for crime and event understanding across law & order, legal activity, governance, and social-issue contexts.

Each record represents a real-world headline paired with four aligned labels capturing: Event type Location Participants Action / outcome

The dataset is designed for NLP research, media analytics, and policy-oriented analysis in low-resource languages.

Note: The dataset reflects news reporting, not legal judgments or verified outcomes.

Data Source & Curation: a. Headlines are collected from publicly available Bengali news sources b. All labels are manually standardized into consistent label spaces c. Headlines: Bengali d. Labels: English (for broader usability and cross-lingual research)

Dataset Structure: Total records: 1K–10K headlines Granularity: One headline = one primary event

Fields per record: Field Description Title Bengali news headline (UTF-8) EventType High-level category (Violence / Legal Activity / etc.) Location City or region tag Participants, Actor, group Action, Procedural state or outcome

Key Features:

  1. Multi-head labels: Enables joint or hierarchical modeling
  2. Low-resource friendly: Bengali text with English labels
  3. Extendable geography: Can be linked with gazetteers

Potential Use Cases: a. Text Classification: Single- or multi-head prediction b. Information Extraction: Actor/action inference from headlines c. Trend Analysis: City-wise or actor-wise event patterns d. Policy & Media Studies: Aggregate reporting insights

Recommended Splits & Metrics: Split: Stratified 70/15/15 or 70/10/20

Metrics:

  1. Macro-F1 (primary)
  2. Micro-F1 / Accuracy (secondary)

Imbalance handling:

  1. Class weights
  2. Focal loss
  3. Balanced sampling

License: This dataset is released under CC BY 4.0. Free to use, share, and adapt for research and educational purposes. Please use responsibly and avoid individual-level attribution or predictive policing.

Citation: @inproceedings{hossain2025masknet, title={MaskNet: Enhancing Crime Event Detection with Feature Masking and Dynamic Attention}, author={Hossain, M. M. and Hossain, M. S. and Chaki, S. and Rahman, M. S. and Ali, A. S.}, booktitle={Proceedings of the 2nd International Conference on Next-Generation Computing, IoT and Machine Learning (NCIM)}, year={2025}, publisher={IEEE} }

GitHub Repository: https://github.com/MIthun667/Crime-Event-Detections/tree/main