ky1916/hatcorpus-bert-authorship

HATCorpus BERT Authorship Classifier

Model Description

This model is a BERT-based binary text classifier trained to distinguish between human-written and AI-generated English text.

• Base model: bert-base-uncased
• Task: Binary text classification
• Labels:
  • 0 → Human-written
  • 1 → AI-generated

The model was trained on HATCorpus, a curated dataset of human and AI-authored text.

---

Intended Use

This model is intended for:

• Research on AI-generated text detection
• Benchmarking authorship classifiers
• Educational and exploratory use

It is not intended for:

• Surveillance or enforcement
• Determining individual authorship
• High-stakes automated decisions

---

Training Data

• Dataset: HATCorpus
• Sources: Wikipedia, Project Gutenberg, AI-generated text
• Language: English
• Split: Train / Validation

For dataset details, see:
👉 https://huggingface.co/datasets/ky1916/HATCorpus

---

Training Details

• Architecture: BERT-base
• Optimizer: AdamW
• Loss: Cross-entropy
• Max sequence length: 512
• Batch size: 8
• Epochs: 3

---

Evaluation

Metric	Value
Accuracy	93.58%
Precision	88.04%
Recall	96.43%
F1-score	92.05%

(Evaluated against dataset at https://www.kaggle.com/datasets/shanegerami/ai-vs-human-text)

---

Usage

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

tokenizer = AutoTokenizer.from_pretrained("your-username/hatcorpus-bert-authorship")
model = AutoModelForSequenceClassification.from_pretrained("your-username/hatcorpus-bert-authorship")

text = "This is an example sentence."
inputs = tokenizer(text, return_tensors="pt", truncation=True)
outputs = model(**inputs)
prediction = torch.argmax(outputs.logits, dim=1).item()

print("AI-generated" if prediction == 1 else "Human-written")

Limitations

• Performance may degrade on very short text
• Model may rely on stylistic cues rather than semantic understanding
• Does not generalize to all LLMs or writing styles

Citation

If you use this model in your research, please cite both the model and dataset:

@model{hatcorpus_bert_2025,
  title     = {HATCorpus BERT Authorship Classifier},
  year      = {2025},
  publisher = {Hugging Face},
  note      = {Fine-tuned from bert-base-uncased for human vs AI text classification}
}