|
|
--- |
|
|
library_name: transformers |
|
|
tags: |
|
|
- sentiment-analysis |
|
|
- lora |
|
|
- roberta |
|
|
- fine-tuned |
|
|
- insurance |
|
|
--- |
|
|
|
|
|
# Model Card for RoBERTa LoRA Fine-Tuned for Insurance Review Rating |
|
|
|
|
|
This model is a fine-tuned version of RoBERTa (`roberta-large`) using LoRA adapters. It is specifically designed to classify English insurance reviews and assign a rating (on a scale of 1 to 5). |
|
|
|
|
|
## Model Details |
|
|
|
|
|
### Model Description |
|
|
|
|
|
This model uses RoBERTa (`roberta-large`) as its base architecture and was fine-tuned using Low-Rank Adaptation (LoRA) to adapt efficiently to the task of insurance review classification. The model predicts a rating from 1 to 5 based on the sentiment and context of a given review. LoRA fine-tuning reduces memory overhead and enables faster training compared to full fine-tuning. |
|
|
|
|
|
- **Developed by:** Lapujpuj |
|
|
- **Finetuned from model:** RoBERTa (`roberta-large`) |
|
|
- **Language(s) (NLP):** English |
|
|
- **License:** Apache-2.0 |
|
|
- **LoRA Configuration:** |
|
|
- Rank (r): 2 |
|
|
- LoRA Alpha: 16 |
|
|
- LoRA Dropout: 0.1 |
|
|
- **Task:** Sentiment-based rating prediction for insurance reviews |
|
|
|
|
|
### Model Sources |
|
|
|
|
|
- **Repository:** [pujpuj/roberta-lora-token-classification](https://huggingface.co/pujpuj/roberta-lora-token-classification) |
|
|
- **Demo:** N/A |
|
|
|
|
|
--- |
|
|
|
|
|
## Uses |
|
|
|
|
|
### Direct Use |
|
|
|
|
|
This model can be directly used to assign a sentiment-based rating to insurance reviews. Input text is expected to be a sentence or paragraph in English. |
|
|
|
|
|
### Downstream Use |
|
|
|
|
|
The model can be used as a building block for larger applications, such as customer feedback analysis, satisfaction prediction, or insurance service improvement. |
|
|
|
|
|
### Out-of-Scope Use |
|
|
|
|
|
- The model is not designed for reviews in languages other than English. |
|
|
- It may not generalize well to domains outside of insurance-related reviews. |
|
|
- Avoid using the model for biased or malicious predictions. |
|
|
|
|
|
--- |
|
|
|
|
|
## Bias, Risks, and Limitations |
|
|
|
|
|
### Bias |
|
|
|
|
|
- The model is trained on a specific dataset of insurance reviews, which might include biases present in the training data (e.g., skewed ratings, linguistic or cultural biases). |
|
|
|
|
|
### Risks |
|
|
|
|
|
- Predictions might not generalize well to other domains or review styles. |
|
|
- Inconsistent predictions may occur for ambiguous or mixed reviews. |
|
|
|
|
|
### Recommendations |
|
|
|
|
|
- Always validate model outputs before making decisions. |
|
|
- Use the model in conjunction with other tools for a more comprehensive analysis. |
|
|
|
|
|
--- |
|
|
|
|
|
## How to Get Started with the Model |
|
|
|
|
|
You can use the model with the following code snippet: |
|
|
|
|
|
```python |
|
|
from transformers import AutoTokenizer |
|
|
from peft import PeftModel |
|
|
|
|
|
# Load tokenizer and model |
|
|
tokenizer = AutoTokenizer.from_pretrained("roberta-large") |
|
|
base_model = AutoModelForSequenceClassification.from_pretrained("roberta-large", num_labels=5) |
|
|
model = PeftModel.from_pretrained(base_model, "pujpuj/roberta-lora-token-classification") |
|
|
|
|
|
# Example prediction |
|
|
review = "The insurance service was quick and reliable." |
|
|
inputs = tokenizer(review, return_tensors="pt", truncation=True, padding=True) |
|
|
outputs = model(**inputs) |
|
|
rating = torch.argmax(outputs.logits, dim=1).item() + 1 |
|
|
print(f"Predicted rating: {rating}") |
|
|
|
