Russian-Language Toxicity Dataset with 89,850 Balanced Comments

A Russian-language dataset of 89,850 social media comments, manually annotated for toxicity. The collection is balanced with approximately 44,925 examples each for toxic (1) and neutral (0) classes. It was created by VladimirRH and last updated on Hugging Face in July 2026.

Use Cases

• Train binary text classifiers for toxicity detection based on labeled Russian comments.
• Benchmark model performance on a balanced dataset of toxic and neutral language.
• Analyze linguistic patterns of toxicity in Russian social media discourse.

Strengths

• Dataset contains 89,850 annotated text records.
• Class distribution is balanced with approximately 44,925 examples per class (toxic/neutral).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Data may reflect geographic and source bias inherent to the original comment datasets.

Provenance

Source

Aggregated from datasets including AlexSham/Toxic_Russian_Comments and s-nlp/ru_paradetox_toxicity.

Collection Method

Likely a balanced sampling and merging of existing Russian toxicity datasets.

Freshness

Last updated 2026-07-11 20:13:12; freshness should be verified.

Geography

Primarily Russian-language sources, suggesting a focus on content from Russian-speaking regions.