Russian Inappropriate Messages Classifier
| Property | Value |
|---|---|
| License | CC BY-NC-SA 4.0 |
| Language | Russian |
| Framework | PyTorch, Transformers |
| Performance | 89% Accuracy |
What is russian-inappropriate-messages?
This model represents a specialized approach to content moderation, focusing on detecting inappropriate messages in Russian that could harm a speaker's reputation. Unlike traditional toxicity classifiers, this model identifies content that may be problematic despite not containing explicit toxic or obscene language. It serves as an additional layer of filtering after standard toxicity detection.
Implementation Details
The model utilizes BERT architecture and has been trained on a carefully curated dataset of inappropriate messages. It achieves impressive metrics with 0.89 weighted average F1-score on the test set, demonstrating robust performance in distinguishing between appropriate and inappropriate content.
- Precision: 0.92 for appropriate and 0.80 for inappropriate content
- Recall: 0.93 for appropriate and 0.76 for inappropriate content
- Overall accuracy: 89% across 10,565 test samples
Core Capabilities
- Detection of reputation-damaging content without explicit toxicity
- Classification of messages related to sensitive topics
- Integration with existing content moderation pipelines
- Support for Russian language content analysis
Frequently Asked Questions
Q: What makes this model unique?
This model specializes in detecting subtle inappropriate content that traditional toxicity filters might miss, focusing on reputation damage potential rather than explicit toxic content.
Q: What are the recommended use cases?
The model is ideal for content moderation systems requiring fine-grained inappropriateness detection, particularly in Russian language contexts. It works best as a secondary filter after standard toxicity detection.
