BAED: A Comprehensive Bengali Emotion Dataset with Transformer-based Evaluation

The Bengali Annotated Emotion Dataset (BAED) draws its content from Bengali novels which form the basis for its emotional expression and cultural element identification capabilities that other NLP datasets do not possess. The system divides into seven categories which enable complete multi-class affective state classification through anger, disgust, fear, joy, sadness, surprise and anticipation. Drawing from both dialogue and narration, BAED offers a rich basis for studying human emotions in text.The system has multiple uses in computational linguistics and sentiment and stylistic analysis and cultural and psychological research and dialogue-level emotion detection and benchmarking emotion classification models with potential future applications in newspaper and social media and conversational data.

The dataset is organized into seven balanced classes, each representing a different emotion domain:

Anger:500 data
Disgust:500 data
Fear:500 data
Joy:500 data
Sadness:500 data
Surprise:500 data
Anticipation:500 data

Total Number of Data: 3,500
Language: Bangla
File Format: CSV file