Russian Toxicity Dataset: 90,586 Texts for Binary Classification

A Russian-language dataset for binary classification of texts as toxic or non-toxic. It contains 72,468 training records, balanced 1:1, with 9,059 validation and 9,059 test records, created by author koryl. The dataset was last updated on 2026-07-16.

Use Cases

• Train a binary text classifier based on the provided 'text' and 'label' columns.
• Benchmark toxicity detection algorithms on a balanced Russian-language dataset.
• Fine-tune language models for content moderation tasks using the described training split.

Strengths

• Training set is balanced 1:1 between toxic and non-toxic classes, with 72,468 records.
• Dataset provides distinct validation and test sets, each with 9,059 records.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last updated 2026-07-16 12:07:55; freshness should be verified.

Provenance

Source

huggingface

Collection Method

Toxic comments sourced from 'klamas/russian-toxic'; non-toxic texts from questions (source truncated in description).

Freshness

2026-07-16 12:07:55