Cyberbullying Detection System — Real-Time Toxic Comment Classifier

1. Introduction

The Cyberbullying Detection System Using Comment Analysis is a Python-based AI project that identifies abusive, hateful, or threatening comments on social platforms. It leverages NLP and Machine Learning to understand linguistic context (not just keywords), enabling real-time moderation support and safer online communities. A Flask dashboard shows live classifications and confidence, helping moderators act quickly and consistently.

2. Existing System vs Proposed System

3. Working

1. Data Collection: Labeled comment datasets (Kaggle/Twitter/Reddit).
2. Pre-processing: Clean text, normalize case, remove stop words/emojis.
3. Feature Extraction: TF-IDF or embeddings (Word2Vec/Glove).
4. Model Training: Train NB/SVM/LSTM classifier on labeled data.
5. Real-Time Prediction: API scores new comments instantly.
6. Result Output: Category + confidence, with mod alerts.

4. Technology Stack

• Language: Python
• Libraries: NLTK, Scikit-learn, TensorFlow/Keras, Pandas, NumPy, Flask
• Algorithms: Naïve Bayes, Logistic Regression, SVM, or LSTM
• Dataset: Kaggle Cyberbullying Tweets or similar comment datasets
• Interface: Flask dashboard for live analysis
• Storage: SQLite3/CSV for logs & results

5. Modules

6. Advantages

• Automatic, real-time toxic content detection
• Understands context better than keyword filters
• Reduces moderator workload with higher accuracy
• Improves user safety across platforms
• Supports multilingual detection with proper training

7. Applications

• Social media and community platforms
• School/college forums and LMS portals
• NGO/government anti-cyberbullying initiatives
• Chat/gaming platforms and discussion boards

Python Integration Sketch (Flask + NLP + ML)

# 1) Preprocess & vectorize
def preprocess(txt):
    txt = normalize(txt)            # lowercase, punctuation, emoji removal
    tokens = tokenize(txt)          # NLTK/regex
    tokens = drop_stopwords(tokens) # NLTK stopwords
    return " ".join(tokens)

X_train = vectorizer.fit_transform(preprocess(t) for t in train_texts)
model   = train_clf(X_train, y_train)  # NB/SVM or build LSTM pipeline

# 2) Realtime API
@app.post("/predict")
def predict():
    txt = request.json["comment"]
    x   = vectorizer.transform([preprocess(txt)])
    y   = model.predict(x)[0]
    p   = model.predict_proba(x).max()
    return {"category": y, "confidence": float(p)}

# 3) Dashboard stream
# Moderators see flagged comments, filter by category/confidence, export reports.
              

What You Get