Text Classifier v2: AI vs SLW (Second Language Writers)

This project is an extension of Text Classifier v1, using a stronger XGBoost model and larger dataset. It explores linguistic data science and syntactic complexity modeling.

Live App

---

Improvements over v1

• Upgraded from Random Forest to XGBoost
• More dataset for training (from 300 → 1000 samples)
• More stable predictions and improved feature interpretability

---

About the syntactic Complexity Indices

The classifier uses L2SCA indices by TAASSC.

Predictions are supported by SHAP contribution plots, showing how each feature influences the outcome toward AI or SLW.

---

Data Overview

The dataset used for model training consists of 1,000 writing samples (500 human, 500 AI):

• Human-written:
  500 essays by second language writers (SLW), sourced from ICNALE
• AI-generated:
  500 essays generated by large language models (LLMs), sourced from LLM-generated Essay Dataset

Data preprocessing by TAASSC.

---

Data Usage Notice

• The .txt files in txt_samples/ are included only for demonstration and learning purposes.
  They are not licensed for reuse, redistribution, or commercial use.

• The dataset file X_binary.csv is private and is not licensed for reuse, redistribution, or modification.
  It is shared solely for demonstration purposes and should not be used for any other purpose.