Social media discourse on feminism: A dataset for sentiment analysis in bangla comments

TL;DR

This paper introduces a dataset of Bangla social media comments about feminism to study public sentiment and abusive language in Bangladesh.

Contribution

The novelty lies in providing a manually annotated Bangla dataset for feminist discourse, focusing on sentiment and abusive language.

Findings

The dataset contains 6,830 Bangla comments collected from Facebook, Instagram, and Twitter.

Comments were annotated for sentiment (neutral, negative, positive) and abusive language.

The dataset supports machine learning tasks like sentiment analysis and gender-based violence studies in Bangla.

Abstract

Bangladesh is a socio-culturally diverse country where perspectives on women’s freedom vary significantly. Social media sites are now important places for sharing feminist ideas through public comments and posts. This study offers a detailed collection of 6,830 comments in Bangla about feminism to help analyze public opinion. Most of this data comes from Facebook, with some also from Instagram and Twitter. Data collection involved systematic extraction from public groups and targeted hashtag searches, including (women’s rights). Native Bangla speakers meticulously annotated each comment by hand to ensure that it was topical and to identify any abusive language. This manual validation procedure guarantees a high-quality dataset appropriate for the study of online gender-based violence in the Bangla language context, sentiment analysis, and abusive language analysis, among other machine learning and NLP tasks. In addition, comments were divided into three sentiment classes: neutral, negative, and positive. This allowed for thorough analysis of feminist discourse on Bangladeshi social media and supervised learning.

Figures