Processed Bangla Social Media comments Dataset (PBMedia)

Collected data was labelled into 10 topic class (others, abusive, political, religious, international, education, food, sports, technology, threat) and 4 sentiment class (positive, negative, neutral, hate . The dataset were annotated by native Bengali speakers.

The dataset is filtered, tokenized, stemmed and concatenated again as a sentence. The dataset supports multiple downstream tasks such as topic classification, sentiment analysis, and the development of real-world automatic censorship models.