DistilCamemBERT-Sentiment
| Property | Value |
|---|---|
| Author | cmarkea |
| Task | French Sentiment Analysis |
| Training Data | 440,509 reviews (Amazon + Allocine) |
| Accuracy | 61.01% (exact), 88.80% (top-2) |
| Paper | HAL Archive |
What is distilcamembert-base-sentiment?
DistilCamemBERT-Sentiment is a specialized French language model fine-tuned for sentiment analysis. Built on DistilCamemBERT architecture, it provides efficient sentiment classification while maintaining robust performance. The model analyzes text across five sentiment categories, from 1 star (terrible) to 5 stars (excellent).
Implementation Details
The model utilizes a distilled version of CamemBERT, achieving significantly faster inference times (95.56ms compared to CamemBERT's 329.74ms) while maintaining competitive accuracy. It was trained on a diverse dataset combining Amazon reviews and Allocine.fr critics, totaling over 440,000 samples.
- Balanced training data from two sources to minimize domain bias
- Optimized for production deployment with reduced inference costs
- Supports both exact and top-2 accuracy metrics
- Compatible with Hugging Face Transformers and ONNX runtime
Core Capabilities
- Five-class sentiment classification (1-5 stars)
- Fast inference time (95.56ms average)
- High top-2 accuracy (88.80%)
- Specialized for French language text
- Production-ready with ONNX optimization support
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its efficient architecture, being a distilled version of CamemBERT that achieves twice the inference speed while maintaining competitive accuracy. It's specifically optimized for French sentiment analysis and trained on a diverse dataset to ensure robust performance across different text styles.
Q: What are the recommended use cases?
The model is ideal for production environments where processing French text sentiment is required at scale. It's particularly suitable for analyzing customer reviews, social media sentiment, and general opinion mining in French text, especially when processing speed is crucial.
