Politically Aggressive Bengali Dataset

The Politically Aggressive Bengali Dataset (PABD) is a manually annotated dataset developed to facilitate the analysis of political aggression in Bengali text. It consists of 2,901 political texts collected from publicly available online sources, specifically Facebook and YouTube, focusing on content from news media pages, political parties, and political commentators.

The data acquisition process involved collecting user-generated textual content representing diverse political opinions and discourse. The dataset includes both politically aggressive and non-aggressive texts. Prior to annotation, the collected data were preprocessed to remove noise, redundancies, and inconsistencies, ensuring improved data quality and usability.

A two-level hierarchical annotation framework was applied. At Level 1, each text is categorized as either politically aggressive (PoAG) or politically non-aggressive (PoNoAG). At Level 2, texts identified as PoAG are further classified into three fine-grained categories: inciting assassination (InASS), inciting communal violence (InCOM), and inciting rebellion (InREB).

The annotation was performed manually by multiple annotators following predefined guidelines to ensure consistency and reliability. Annotation quality was assessed using standard inter-annotator agreement measures.

The dataset is organized in structured formats, including raw data, preprocessed text, and annotated labels, with unique identifiers assigned to each instance. PABD can support research in natural language processing, hate speech detection, political discourse analysis, and low-resource language modeling, particularly for the Bengali language.