Detecting Spam YouTube Comments

Overview

YouTube’s immense popularity—with over 2.7 billion monthly users—has attracted spammers and bots that post irrelevant or promotional comments, degrading user experience and engagement. This project applies natural language processing and machine learning to automatically detect and filter spam comments, achieving high accuracy on a Kaggle benchmark dataset. :contentReference[oaicite:0]{index=0}:contentReference[oaicite:1]{index=1}

Data

• Source: Kaggle “Detecting Spam YouTube Comments” dataset
• Samples: 1,369 comments
• Features used: CONTENT (comment text) and CLASS (spam vs. not spam) :contentReference[oaicite:2]{index=2}:contentReference[oaicite:3]{index=3}
• Class distribution: ~51.9 % spam, ~48.1 % not spam :contentReference[oaicite:4]{index=4}:contentReference[oaicite:5]{index=5}
• Comment length: 3–1,200 characters (mean ≈ 96)
• Vocabulary size after TF-IDF: 884 features (with min_df=2, default max_df=1) :contentReference[oaicite:6]{index=6}:contentReference[oaicite:7]{index=7}

Methods

1. Preprocessing
   • Lowercasing, removal of non-alphabetic characters, digits, and extra whitespace
   • Stop-word filtering using English stop-word list :contentReference[oaicite:8]{index=8}:contentReference[oaicite:9]{index=9}
2. Feature Extraction
   • TF-IDF vectorization (chosen over Bag-of-Words and Word2Vec for short-text suitability) :contentReference[oaicite:10]{index=10}:contentReference[oaicite:11]{index=11}
3. Modeling
   • Logistic Regression and Support Vector Machine (SVM) compared via 5-fold cross-validation
   • Outlier analysis using Z-score and Isolation Forest :contentReference[oaicite:12]{index=12}:contentReference[oaicite:13]{index=13}
   • Hyperparameter tuning with GridSearchCV
4. Evaluation Metrics
   • Accuracy, F1-score, and ROC AUC

Results

• Final model: SVM with RBF kernel (C=1, gamma=0.5)
• Kaggle test accuracy: 93.66 % (ranked 27th) :contentReference[oaicite:14]{index=14}:contentReference[oaicite:15]{index=15}
• Test set performance:
  • Accuracy: 94.7 %
  • F1 Score: 93.33 %
  • AUC: 0.9842 :contentReference[oaicite:16]{index=16}:contentReference[oaicite:17]{index=17}

Repository Structure

├── Dataset/ # Raw and output CSV files ├── Scripts/ # Model training and evaluation scripts │ └── classifier.py # Main classifier implementation ├── README.md # Project summary and instructions └── requirements.txt # Python dependencies

Installation

1. Clone the repository:

   git clone github.com/WilliamKwanProgramming/sta314_finalproject.git
   cd sta314_finalproject

2. Install dependencies:

   pip install -r requirements.txt

Usage

1. Prepare training and test CSV files under Dataset/.

2. Run the classifier script:

   python Scripts/classifier.py

3. Output predictions will be saved to Dataset/output/.

Future Work

• Incorporate ensemble tree methods (e.g., Random Forest, Gradient Boosting)
• Expand training data to include diverse video categories
• Enhance robustness to misspellings and semantic variations

References

• Kaggle dataset “Detecting Spam YouTube Comments”
• Word2Vec pre-trained vectors (Google News)
• Scikit-learn, NLTK, Gensim documentation