Sentiment Analysis is the task of classifying polarity of a given text.
SentNoB: A Dataset for Analysing Sentiment on Noisy Bangla Texts is a dataset which contains 15,728 instances with each labeled with one of three-class labels. This work also proposes, unique word percentage, a new evaluation metric for datasets. Models are evaluated based on micro-averaged F1 score.
| Model | F1-score | Paper / Source | Code |
|---|---|---|---|
| U + B + T + C2 + C3 + C4 + C5 | 64.61 | SentNoB: A Dataset for Analysing Sentiment on Noisy Bangla Texts | Official |