This dataset consists of 50,016 samples of comments extracted from Algerian YouTube channels. It is manually annotated with 3 classes (the label column) and is not balanced. Here are the number of rows of each class:
0 (Negative): 17,033 (34.06%)
1 (Neutral): 11,136 (22.26%)
2 (Positive): 21,847 (43.68%)
Please note that there are some swear words in the dataset, so please use it with caution.