SaherMuhamed/bert-ai-detector-coling-finetuned

BERT Fine-tuned Model for AI Content Detection 👾

This directory contains a fine-tuned BERT model for detecting AI-generated content.

Model Overview

• Base Model: BERT-base-uncased (110M parameters)
• Task: Binary classification (AI-generated vs Human-written text)
• Max Sequence Length: 256 tokens
• Fine-tuned on: COLING 2025 MGT Dataset

HuggingFace Hub Usage

You can use our model directly via HuggingFace Hub:

from transformers import pipeline

# Load the model
classifier = pipeline("text-classification", model="SaherMuhamed/bert-ai-detector-coling-finetuned")

# Example text
text = "Your text to classify here"

# Get prediction
result = classifier(text)

# Print result
print(f"Label: {result[0]['label']}")
print(f"Confidence: {result[0]['score']:.2%}")

Example output:

Label: AI_GENERATED
Confidence: 92.45%

Model Performance

The model uses confidence thresholding (0.6) for more reliable predictions, with the following features:

• Handles texts of any length (automatically truncates to 256 tokens)
• Returns probability scores for both classes
• GPU-compatible with fallback to CPU

Dependencies

• tensorflow
• transformers
• numpy
• torch