Practika Fiction Support Ru: Russian Customer Support Toxicity Classification Dataset

Practika Fiction Support Ru is a dataset for training a toxicity classifier for user messages to the "KinoPotok" chat support. It contains 120 total examples, with a balanced 50/50 split between toxic and non-toxic classes. The dataset was created by author pogram1st and was last updated on July 15, 2026.

Use Cases

• Train a binary classifier to detect toxic messages in Russian customer support chats based on the described label categories.
• Benchmark model performance on a small, balanced dataset for toxicity detection tasks.
• Analyze patterns in toxic Russian user messages, such as insults, threats, or passive aggression, as outlined in the description.

Strengths

• Class distribution is explicitly balanced 50/50 between toxic and non-toxic labels.
• Toxic label categories are specifically defined, including insults, threats, profanity, passive aggression, and personal attacks.
• Dataset size is specified as 120 total examples, with 96 for training and 24 for testing.

Limitations

• The dataset is described as small, with only 120 total examples, which may limit model generalization.
• Column-level documentation is absent; field semantics must be inferred after download.
• Last updated 2026-07-15 09:21:17; freshness should be verified.

Provenance

Source

huggingface, author pogram1st

Collection Method

Created for training a toxicity classifier for the "KinoPotok" support chat.

Freshness

Last updated 2026-07-15 09:21:17.