KLUE RoBERTa Small 3i4k Intent Classification
| Property | Value |
|---|---|
| Parameter Count | 68.1M |
| License | CC-BY-NC-4.0 |
| Base Model | KLUE/RoBERTa-small |
| Training Dataset | 3i4k (61,255 samples) |
| Accuracy | 90% (weighted avg) |
What is klue-roberta-small-3i4k-intent-classification?
This is a specialized Korean language model fine-tuned for intent classification tasks. Based on KLUE's RoBERTa-small architecture, it's been specifically trained to identify seven different types of utterance intents in Korean text, including statements, questions, commands, and rhetorical variations.
Implementation Details
The model was fine-tuned using the 3i4k dataset, comprising 46,863 training samples, 8,271 validation samples, and 6,121 test samples. Training utilized Adam optimizer with a learning rate of 5e-05 and batch size of 32, implementing early stopping for optimal performance.
- Seven distinct intent classifications: fragment, statement, question, command, rhetorical question, rhetorical command, and intonation-dependent utterance
- Achieves impressive F1-scores ranging from 0.62 to 0.97 across different intent categories
- Particularly strong in classifying fragments (0.97 F1) and questions (0.96 F1)
Core Capabilities
- High-accuracy intent classification for Korean text
- Robust performance across various utterance types
- Easy integration with HuggingFace's Transformers library
- Efficient processing with relatively small parameter count
Frequently Asked Questions
Q: What makes this model unique?
The model combines KLUE's RoBERTa architecture with specialized Korean intent classification capabilities, offering high accuracy while maintaining a relatively small parameter count of 68.1M, making it efficient for production deployment.
Q: What are the recommended use cases?
This model is ideal for Korean language processing applications requiring intent classification, such as chatbots, customer service automation, and text analysis systems. It's particularly effective for systems needing to distinguish between different types of utterances like questions, commands, and statements.
