Polish Twitter Emotions Dataset with 8 Labels, Refined

Twitteremo Pl Refined is a cleaned and improved version of the original Polish Twitter emotion dataset. The revision involved pruning sparsely-covered labels, detecting and correcting label noise using Cleanlab, and reviewing labels with assistance from large language models. The dataset was created by yazoniak and was last updated on 2025-10-10.

Use Cases

• Train emotion classification models based on the 8 refined labels.
• Benchmark Polish-language NLP models on a cleaned social media text corpus.
• Study label noise correction techniques using the described Cleanlab methodology.
• Analyze the linguistic expression of sarcasm ('sarkazm') in Polish social media text.

Strengths

• Dataset has undergone a multi-step refinement process including label pruning and noise correction.
• An explicit 'sarkazm' (sarcasm) label was added, which is a specific and challenging emotion category.
• The cleaning process leveraged out-of-fold predictions and large language model assistance.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect temporal and demographic biases inherent to Twitter as a source.

Provenance

Source

Original Polish Twitter data, refined by yazoniak.

Collection Method

Data was cleaned using Cleanlab for noise detection and LLM-assisted label review.

Freshness

Last updated 2025-10-10 07:35:06; freshness should be verified.

Geography

Poland (inferred from language)