Russian Toxic Comments for Binary Classification, 130k Balanced Records

Russian-language text data for binary toxicity classification, containing approximately 130,000 records with balanced classes. The dataset was created by user 'ermaksxc' and was last updated on July 16, 2026. It includes pre-defined train, validation, and test splits stratified by class.

Use Cases

• Train binary text classifiers for toxicity detection based on Russian-language comments.
• Benchmark model performance on a balanced Russian toxicity dataset with predefined splits.
• Study linguistic patterns of toxic language in Russian online discourse.
• Fine-tune pre-trained language models for Russian content moderation tasks.

Strengths

• Approximately 130,000 text records provide a substantial sample size for model training.
• Classes are balanced (50/50), which can help mitigate class imbalance issues during training.
• Predefined train (80%), validation (10%), and test (10%) splits are stratified by class.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The source and collection methodology for the text comments are not specified, which may obscure potential biases.
• Last updated 2026-07-16 18:47:42; freshness should be verified as the data may be several years old.

Provenance

Source

huggingface user 'ermaksxc'.

Freshness

Last updated 2026-07-16 18:47:42.

Geography

Text is in Russian, suggesting a primary focus on Russian-speaking regions or online communities.