README.md

metadata

language: en
license: mit
tags:
  - xlnet
  - automatic-short-answer-grading
  - regression
  - education
  - short-answer
  - assessment
  - grading
datasets:
  - Meyerger/ASAG2024
metrics:
  - mse
  - rmse
  - mae
  - pearson correlation
model-index:
  - name: xlnet-regression
    results:
      - task:
          type: regression
          name: automatic short answer grading
        metrics:
          - type: mse
            value: 0.058389
          - type: rmse
            value: 0.241639
          - type: mae
            value: 0.153142
          - type: pearson correlation
            value: 0.80115
pipeline_tag: text-classification

ASAG XLNet Regression Model

This model evaluates student answers by comparing them to reference answers and predicting a grade (regression).

Model Details

• Model Type: XLNet for Regression
• Task: Automatic Short Answer Grading (ASAG)
• Framework: PyTorch/Transformers
• Base Model: xlnet-base-cased

Usage

from transformers import XLNetTokenizer, XLNetForSequenceClassification
import torch

# Load model and tokenizer
tokenizer = XLNetTokenizer.from_pretrained("kenzykhaled/xlnet-regression")
model = XLNetForSequenceClassification.from_pretrained("kenzykhaled/xlnet-regression")

# Prepare inputs
student_answer = "It is vision."
reference_answer = "The stimulus is seeing or hearing the cup fall."

inputs = tokenizer(
    text=student_answer,
    text_pair=reference_answer,
    return_tensors="pt",
    padding=True,
    truncation=True
)

# Get prediction
with torch.no_grad():
    outputs = model(**inputs)

# Get predicted grade (normalized between 0-1)
predicted_grade = outputs.logits.item()
predicted_grade = max(0, min(1, predicted_grade))
print(f"Predicted grade: {predicted_grade:.4f}")

Training Data

This model was trained on the Meyerger/ASAG2024 dataset.

Use Cases

• Automated grading of student short-answer responses
• Educational technology platforms
• Learning management systems
• Assessment tools
• Teacher assistance for grading

Limitations

• The model is trained on specific educational domains and may not generalize well to all subjects
• Performance depends on the similarity of input data to the training data
• Should be used as an assistive tool for grading rather than a complete replacement for human evaluation

Ethical Considerations

When using this model for automated grading:

• Be transparent with students about the use of AI for grading
• Consider potential biases in evaluation
• Provide human review of edge cases
• Allow students to appeal automated grades