Twitter Emotion PL (base) is a Polish-language emotion classification model built on top of herbert-base. It distinguishes between four emotions commonly found in social media: joy, optimism, sadness, and anger.
Trained on a translated version of the TweetEval dataset, it delivers strong performance across all key metrics (F1 macro: 0.756, accuracy: 0.789) and processes over 130 tweets per second on a single RTX 3090 (base version).
This model is well-suited for applications in media monitoring, opinion analysis, and research on online discourse in Polish.

Base: https://huggingface.co/bardsai/twitter-emotion-pl-base

Fast: https://huggingface.co/bardsai/twitter-emotion-pl-fast

How to use

You can use this model directly with a pipeline for text classification:

from transformers import pipeline

nlp = pipeline("text-classification", model="bardsai/twitter-emotion-pl-base")
nlp("Nigdy przegrana nie sprawiła mi takiej radości. Szczęście i Opatrzność mają znaczenie Gratuluje @pzpn_pl")

[{'label': 'joy', 'score': 0.5163766145706177}]

Performance

Base

Fast