The rise of large language models (LLMs) such as GPT-3 has led to an explosion in the creation of AI-generated text. These models can produce highly coherent and contextually relevant text that can easily be mistaken for human writing.

This project focuses on developing a deep learning-based system to distinguish between AI-generated text and human-generated text. By leveraging a comprehensive dataset from Kaggle, which contains labeled examples of both types of text, we aim to build and train a model capable of identifying the origin of the text with high accuracy.

The project is organized into the following directories:

1. Data: Contains the datasets used for training and evaluation.
2. Notebooks: Contains Jupyter notebooks for experimentation and analysis.
3. Results: Stores the results and evaluation metrics obtained from the models.

To run the code in this project, follow these steps:

1. Clone the repository: git clone https://github.com/polonium31/SEP-740-Project_Detecting-AI-vs-Human-generated-text.git

2. The project contains three main folders:

• dataset: Includes the datasets used for training and evaluation.

  • Training_Essay_Data.csv: The main dataset.
  • cleaned_data.csv: The dataset after exploratory data analysis (EDA) is performed.

• notebooks: Contains Jupyter notebooks for experimentation and analysis.

  • Data_Processing.ipynb: Contains EDA code.
  • ML_Code.ipynb: Contains code for solving the problem using Machine Learning techniques.
  • DL_LSTM_model.ipynb: Contains code for using the LSTM model.
  • DL_BERT_model.ipynb: Contains code for using the BERT model.
  • DL_RoBERTa_model.ipynb: Contains code for using the RoBERTa model.
  • DL_GPT2_model.ipynb: Contains code for using the GPT2 model.

• results: Contains all important graphs.

All Deep Learning models are available here: https://drive.google.com/drive/folders/1RAcC2Rt0oylTFr5aONWQrZefPPnl6Qyj?usp=sharing

• Open and run Data_Processing.ipynb to perform exploratory data analysis (EDA) and necessary data preprocessing.
• Open ML_Code.ipynb to explore and execute Machine Learning models like Random Forest, AdaBoost, and XGBoost.
• Open DL_LSTM_model.ipynb to work with the LSTM model for deep learning-based text classification.
• Open DL_BERT_model.ipynb to work with the BERT model for deep learning-based text classification.
• Open DL_RoBERTa_model.ipynb to work with the RoBERTa model for deep learning-based text classification.
• Open DL_GPT2_model.ipynb to work with the GPT2 model for deep learning-based text classification.

The dataset used in this project can be obtained from Kaggle. It is available at the following link: LLM Detect AI-Generated Text Dataset.

To use this dataset, follow these steps:

1. Visit the provided Kaggle link.
2. Click on the "Download" button to download the dataset.
3. Extract the downloaded file to a desired location on your local machine.

• Contributions: EDA, Report
• GitHub Profile Link: Raj Joshi

• Contributions: Initial problem defining, ML
• GitHub Profile Link: Simran Chadda

• Contributions: Initial problem defining, DL (LSTM & BERT), Report
• GitHub Profile Link: Jainish Patel

• Contributions: DL (GPT), Presentation
• GitHub Profile Link: Mayur Patel

1. Term Frequency-Inverse Document Frequency (TF-IDF): Text Vectorization. (n.d.). Towards Data Science. Retrieved from https://towardsdatascience.com/text-vectorization-term-frequency-inverse-document-frequency-tfidf-5a3f9604da6d.
2. Devlin, J., Chang, M. W., Lee, K., & Toutanova, K. (2019). BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. arXiv: https://arxiv.org/abs/1810.04805.
3. Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, Veselin Stoyanov. RoBERTa: A Robustly Optimized BERT Pretraining Approach. arXiv: https://arxiv.org/abs/1907.11692.
4. Weber-Wulff, D., Anohina-Naumeca, A., Bjelobaba, S., Foltýnek, T., Guerrero-Dib, J., Popoola, O., Šigut, P., & Waddington, L. (2023). Testing of detection tools for AI-generated text. https://edintegrity.biomedcentral.com/articles/10.1007/s40979-023-00146-z
5. Islam, N., Sutradhar, D., Noor, H., Raya, J. T., Maisha, M. T., & Farid, D. M. (Year). Distinguishing Human Generated Text From ChatGPT Generated Text Using Machine Learning. arXiv: https://arxiv.org/abs/2306.01761.