AI Generated Text Detection

B.Sc. Computer Science Final Seminar Project.

This project explores methods for distinguishing AI-generated text from human-written text using a progression of Natural Language Processing (NLP) and Machine Learning techniques.

Project Goal

The objective was to investigate whether machine learning models can identify patterns that differentiate AI-generated content from human-authored content and compare the effectiveness of classical and modern NLP approaches.

Dataset

The project uses the Kaggle "AI vs Human Text" dataset and includes:

• Data validation and preprocessing
• Length-based filtering
• Exploratory Data Analysis (EDA)
• Text visualization techniques

Models Evaluated

Baseline Model

• Bag of Words (BoW)
• Multinomial Naive Bayes

Embedding-Based Model

• SentenceTransformers (all-MiniLM-L6-v2)
• XGBoost

Transformer-Based Model

• BERT
• PyTorch
• Hugging Face Transformers

Additional Experiments

• TF-IDF feature filtering
• Named Entity Recognition (NER) using spaCy
• Ensemble ("Mega Model") weighted voting approach
• Model agreement/disagreement analysis

Technologies

• Python
• pandas
• scikit-learn
• matplotlib
• SentenceTransformers
• XGBoost
• PyTorch
• Hugging Face Transformers
• spaCy

Project Structure

• AI_vs_Human_Text.ipynb - Original Google Colab notebook
• AI_vs_Human_Text.py - Python export of the notebook

Authors

Developed as a joint B.Sc. seminar project.

Contributors:

• Ido Davidi
• yvcode