BanglaSentiNet: A Unified Bangla Dataset for Polarity and Emotion Classification

Description

BanglaSentiNet is a corpus of 28,000 Bangla text records jointly annotated for fine-grained sentiment polarity and emotion. Each record contains a five-class polarity label, a continuous polarity score (−1 to +1), seven continuous emotion scores, and a categorical emotion label. Dataset Schema | Field | Type | Description | | --- | --- | --- | | `sl_no` | Integer | Unique record identifier. | | `Bengali` | Text | Annotated Bangla text. | | `polarity_label` | Categorical | Five-class polarity label. | | `polarity_score` | Numeric | Polarity score from −1 to +1 | | `neutral` | Numeric | Neutral emotion score. | | `happy` | Numeric | Happy emotion score. | | `anger` | Numeric | Anger emotion score. | | `disgust` | Numeric | Disgust emotion score. | | `sadness` | Numeric | Sadness emotion score. | | `surprise` | Numeric | Surprise emotion score. | | `fear` | Numeric | Fear emotion score. | | `emotion_label` | Categorical | Dominant emotion label. | Dataset Summary | Characteristic | Description | | --- | --- | | Records | 28,000 | | Language | Bangla | | Polarity classes | 5 | | Emotion categories | 7 | | Polarity representation | Categorical and continuous | | Emotion representation | Continuous scores and categorical label | | Polarity range | −1 to +1 | | Emotion score range | 0 to 1 | | Format | CSV | Annotation Three human annotators independently labelled all records using a common annotation scheme. Final categorical labels were determined by majority vote, with a fourth annotator adjudicating complete disagreements. Continuous polarity and emotion scores were calculated as the mean of the three human annotations.

Institutions

• Khulna Khan Bahadur Ahsanullah University

  Khulna Division, Khulna
• University of Windsor

  Ontario, Windsor

Categories

Licence