Update README.md

README.md

@@ -1,24 +1,28 @@
----
-license: mit
-dataset_info:
-  features:
-  - name: text
-    dtype: string
-  - name: labels
-    dtype: float64
-  - name: label_text
-    dtype: string
-  - name: label
-    dtype: int64
-  splits:
-  - name: train
-    num_bytes: 833857
-    num_examples: 5755
-  download_size: 405380
-  dataset_size: 833857
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
----
+---
+license: mit
+dataset_info:
+  features:
+  - name: text
+    dtype: string
+  - name: labels
+    dtype: float64
+  - name: label_text
+    dtype: string
+  - name: label
+    dtype: int64
+  splits:
+  - name: train
+    num_bytes: 833857
+    num_examples: 5755
+  download_size: 405380
+  dataset_size: 833857
+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 based on the L-HSAB set mentioned in the original. 
+The current dataset keeps the original class (labels) and adds 2 columns: label_text and label. Label text was generated by LLM (Qwen 2.4 14B quantized). 
+As can be seen, neither classification is perfect.