ToxSyn-PT: 53,274 Brazilian Portuguese Sentences for Fine-Grained Hate Speech Detection

ToxSyn-PT is a large-scale synthetic dataset containing 53,274 sentences for hate speech detection in Brazilian Portuguese. The dataset is equally balanced between toxic and non-toxic labels and covers nine legally protected minority groups, including Black, Women, LGBTQIA+, and Elderly. It was created by AKCIT and last updated on the platform in June 2026.

Use Cases

• Fine-grained hate speech classification based on the nine protected minority groups mentioned in the description.
• Training models to distinguish toxic mentions from benign counterexamples, as the description notes systematic inclusion of non-toxic examples.
• Benchmarking synthetic data generation techniques for NLP tasks in low-resource languages like Brazilian Portuguese.

Strengths

• Contains 53,274 sentences, providing a substantial corpus for model training.
• Explicitly balanced with equal numbers of toxic and non-toxic labels, which can help mitigate class imbalance.
• Covers nine specific legally protected groups, enabling fine-grained analysis of hate speech targets.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known, but other structural details like file formats and specific column names are unknown.
• Data is synthetic, which may limit its realism and applicability to real-world social media content.

Provenance

Source

huggingface, author AKCIT

Collection Method

Synthetically generated, as described.

Freshness

Last updated 2026-06-04 02:19:03; freshness should be verified.

Geography

Likely focused on Brazilian Portuguese language contexts.