---
license: mit
dataset_info:
  features:
  - name: text
    dtype: string
  - name: labels
    dtype: int64
  - name: label_text
    dtype: string
  - name: label
    dtype: int64
  - name: dataset
    dtype: string
  splits:
  - name: train
    num_bytes: 2605687
    num_examples: 11546
  download_size: 1306400
  dataset_size: 2605687
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---
This __hate speeh__ dataset is derived from [manueltonneau/arabic-hate-speech-superset](https://huggingface.co/datasets/manueltonneau/arabic-hate-speech-superset). 
The original dataset has loading issues. This dataset is mainly based on the __L-HSAB__ and __alsafari__ sets mentioned in the original. 
Around 3000 other records from other datasets were added for better (not full) coverage.  
The current dataset keeps the original class (labels - apparently not as accurate as should be) and adds 2 columns: label_text and label (0: negative, 1:positive). 
Label text was first generated by LLM (Qwen 2.5 14B quantized). Label was fully revised after that and reflects my personal opinion (not an exat science).  

A [hate speech classifier model](https://huggingface.co/akhooli/setfit_ar_hs) based on this dataset is available.  

* Note: this dataset was updated on Nov. 14, 2024. It has around 11500 entries all manually inspected and labeled
  (my personal assessment, so not always correct or as intended).  
* __Disclaimer__: some entries contain obscene language, swearing and other forms of unacceptable terminology
 (data taken as is, removed only instances of "@user" from the original). 
The purpose of the dataset is to train a classifier and not to publish or endorse accpeptable/unacceptable content.
* [LinkedIn article](https://www.linkedin.com/posts/akhooli_arabic-hate-speech-detection-is-not-an-easy-activity-7261021456023609344-UJzM)