# 🧬 Semantic Knowledge Driven Opinion Mining and Classification of COVID-19 Vaccination Data > **Major Academic Project** — Hybrid AI System: Supervised Machine Learning + NLP Sentiment Analysis + Rule-Based Semantic Knowledge Engine --- ## 📐 ASCII Architecture Diagram ``` ╔══════════════════════════════════════════════════════════════════════════════╗ ║ SEMANTIC KNOWLEDGE DRIVEN COVID-19 OPINION MINING SYSTEM ║ ╠══════════════════════════════════════════════════════════════════════════════╣ ║ ║ ║ ┌──────────────────────────────────────────────────────────┐ ║ ║ │ USER INPUT (React.js) │ ║ ║ │ Age | Symptoms Score(1-10) | Vacc.Status | Opinion Text │ ║ ║ └──────────────────────────┬───────────────────────────────┘ ║ ║ │ HTTP POST /api/analyze ║ ║ ▼ ║ ║ ┌──────────────────────────────────────────────────────────┐ ║ ║ │ FLASK REST API (Orchestration Layer) │ ║ ║ │ Python 3 · Port 5000 │ ║ ║ └──────────┬──────────────────┬──────────────────┬─────────┘ ║ ║ │ │ │ ║ ║ ▼ ▼ ▼ ║ ║ ┌──────────────────┐ ┌────────────────┐ ┌────────────────────────┐ ║ ║ │ MODULE 1: ML │ │ MODULE 2: NLP │ │ MODULE 3: KNOWLEDGE │ ║ ║ │ Random Forest │ │ VADER Engine │ │ RULE-BASED INFERENCE │ ║ ║ │ Classifier │ │ │ │ ENGINE (Semantic Layer)│ ║ ║ │ │ │ Compound Score│ │ │ ║ ║ │ Inputs: │ │ Pos/Neu/Neg │ │ IF-THEN Production │ ║ ║ │ · Age │ │ Polarity │ │ Rules: │ ║ ║ │ · Symptoms │ │ Label │ │ R1: Pos+Unvacc+High→ │ ║ ║ │ · Vaccinated │ │ │ │ HIGH RISK │ ║ ║ │ │ │ Input: │ │ R2: Pos+Vacc→ │ ║ ║ │ Output: │ │ Opinion Text │ │ MEDIUM RISK │ ║ ║ │ · COVID +/- │ │ │ │ R3: Pos+Unvacc→ │ ║ ║ │ · Confidence % │ │ │ │ HIGH RISK │ ║ ║ │ · Feature Imp. │ │ │ │ R4: Neg+StrongNegSent→│ ║ ║ └────────┬─────────┘ └───────┬────────┘ │ ADVISORY RISK │ ║ ║ │ │ │ R5: Neg→ LOW RISK │ ║ ║ └───────────────────┴──────────┴──────────┬────────────┘ ║ ║ │ ║ ║ ▼ ║ ║ ┌─────────────────────────────────────────────────────────────┐ ║ ║ │ MODULE 4: TREATMENT KNOWLEDGE BASE │ ║ ║ │ HIGH RISK → Immediate Medical Consultation (CRITICAL) │ ║ ║ │ MEDIUM RISK → Home Isolation + Medication (MODERATE) │ ║ ║ │ LOW RISK → Preventive Monitoring (LOW) │ ║ ║ │ ADVISORY RISK → Counseling + Awareness (ADVISORY) │ ║ ║ └─────────────────────────────────────────────────────────────┘ ║ ║ │ ║ ║ ▼ ║ ║ ┌──────────────────────────────────────────────────────────┐ ║ ║ │ OUTPUT DISPLAY (React.js) │ ║ ║ │ COVID Prediction | Sentiment Score | Risk | Treatment │ ║ ║ └──────────────────────────────────────────────────────────┘ ║ ╚══════════════════════════════════════════════════════════════════════════════╝ ``` --- ## 📁 Project Folder Structure ``` major/ ├── backend/ │ ├── app.py ← Flask API + ML + NLP + Knowledge Engine │ ├── requirements.txt ← Python dependencies │ ├── test_api.py ← API test script │ └── venv/ ← Python virtual environment │ └── frontend/ ├── public/ │ └── index.html ← HTML template ├── src/ │ ├── App.js ← Main React component (all UI) │ ├── index.css ← Global styling (dark academic theme) │ └── index.js ← React entry point ├── package.json └── ... ``` --- ## 🚀 How to Run ### Step 1 — Start the Backend (Flask) ```bash cd major/backend # (Only first time) Create virtual environment & install: python -m venv venv .\venv\Scripts\pip install -r requirements.txt # Windows # OR: source venv/bin/activate && pip install -r requirements.txt (Linux/Mac) # Start Flask server: .\venv\Scripts\python app.py ``` Backend runs at: `http://localhost:5000` --- ### Step 2 — Start the Frontend (React) ```bash cd major/frontend npm install # (only first time) npm start ``` Frontend runs at: `http://localhost:3000` --- ## 🧪 Example Test Inputs & Expected Outputs ### Test Case 1: HIGH RISK | Field | Value | |-----------------|-----------------------------------------------| | Age | 68 | | Symptoms Score | 9 | | Vaccinated | No | | Opinion Text | "I refuse vaccines. They are dangerous." | **Expected Output:** - COVID Status: **Positive** - Sentiment: **Negative** (compound ≈ -0.65) - Risk Level: **High Risk** (Rule R1 fired) - Treatment: **Immediate Medical Consultation** (CRITICAL) --- ### Test Case 2: MEDIUM RISK | Field | Value | |-----------------|-----------------------------------------------------------| | Age | 45 | | Symptoms Score | 6 | | Vaccinated | Yes | | Opinion Text | "I got vaccinated and feel somewhat protected." | **Expected Output:** - COVID Status: **Positive** - Sentiment: **Positive** (compound > 0.05) - Risk Level: **Medium Risk** (Rule R2 fired) - Treatment: **Home Isolation + Medication** (MODERATE) --- ### Test Case 3: LOW RISK | Field | Value | |-----------------|--------------------------------------------------------| | Age | 28 | | Symptoms Score | 2 | | Vaccinated | Yes | | Opinion Text | "Vaccines are amazing! I love and support them." | **Expected Output:** - COVID Status: **Negative** - Sentiment: **Positive** (compound ≈ +0.78) - Risk Level: **Low Risk** (Rule R5 fired) - Treatment: **Preventive Monitoring** (LOW) --- ### Test Case 4: ADVISORY RISK | Field | Value | |-----------------|-----------------------------------------------------------| | Age | 32 | | Symptoms Score | 2 | | Vaccinated | No | | Opinion Text | "These vaccines are terrible and dangerous. I oppose them."| **Expected Output:** - COVID Status: **Negative** - Sentiment: **Negative** (compound ≤ -0.40) - Risk Level: **Advisory Risk** (Rule R4 fired) - Treatment: **Counseling + Vaccine Awareness** (ADVISORY) --- ## 🎓 Academic Explanation for Viva ### Q1: Why is this "Semantic Knowledge-Driven"? **Answer:** The system is *semantic* because it doesn't just match patterns — it understands the *meaning* and *relationships* between domain concepts. The Knowledge Engine encodes medical domain expert knowledge as structured Production Rules in the form: ``` IF AND THEN ``` For example: ``` R1: IF COVID_Positive AND Unvaccinated AND HighSymptoms → HighRisk ``` This is semantically meaningful because each entity (`COVID_Positive`, `Unvaccinated`, `HighSymptoms`) represents a real-world medical concept with established clinical relationships. The engine uses **forward chaining** to derive new facts (risk levels) from known facts — exactly analogous to how an **OWL reasoner** or **knowledge graph** would infer new triples. --- ### Q2: Why is this called a Hybrid Architecture? **Answer:** The system combines **three distinct AI paradigms**: | Paradigm | Module | Role | |----------|--------|------| | **Statistical/Subsymbolic ML** | RandomForest Classifier | Learns patterns from data; handles numerical feature relationships | | **Lexicon-based NLP** | VADER Sentiment Analysis | Processes free text using a curated sentiment lexicon | | **Symbolic AI / Knowledge Reasoning** | Rule-Based Engine | Encodes expert knowledge as explicit IF-THEN logical rules | This hybrid approach overcomes the limitations of each individual approach: - ML alone cannot reason about unvaccinated public health risk - NLP alone cannot determine COVID infection probability - Rules alone cannot generalize from patient data --- ### Q3: Why RandomForest specifically? **Answer:** RandomForest was chosen because: 1. It is an **ensemble learner** (combines 150 decision trees) — reducing overfitting 2. It handles **non-linear relationships** between age, symptoms, and vaccination 3. It provides **feature importance metrics** for model interpretability 4. It is robust to **small training sets** and **class imbalance** (`class_weight='balanced'`) 5. It is industry-standard for **tabular clinical data** (used in healthcare AI research) --- ### Q4: Why VADER over BERT/GPT for sentiment? **Answer:** 1. VADER is **purpose-built for short, opinionated text** (similar to tweets/survey responses) 2. It handles **ALL CAPS, punctuation emphasis** (!!!), and **degree modifiers** ("very", "extremely") 3. It requires **no training data** and works out-of-the-box for medical opinion text 4. It is **computationally lightweight** — real-time analysis without GPU 5. It is **widely cited in COVID-19 vaccine sentiment literature** (Lin et al., 2021) --- ### Q5: What is the "working memory" in the Knowledge Engine? **Answer:** In production rule systems, the **Working Memory** holds all currently known facts about the patient at inference time: ```python { 'covid_positive': True, 'vaccinated': False, 'high_symptoms': True, # symptoms_score >= 6 'strong_negative_sentiment': False # compound <= -0.40 } ``` The engine pattern-matches these facts against rule conditions using **forward chaining** — incrementally firing rules until a conclusion is reached. This is the same mechanism used in expert systems like MYCIN and CLIPS. --- ## 📦 Dependencies (requirements.txt) | Package | Version | Purpose | |---------|---------|---------| | flask | 3.0.3 | REST API web framework | | flask-cors | 4.0.1 | Cross-origin resource sharing (React↔Flask) | | scikit-learn | 1.4.2 | RandomForest ML classifier | | vaderSentiment | 3.3.2 | VADER sentiment analysis | | numpy | 1.26.4 | Numerical computation | | pandas | 2.2.2 | Data manipulation | | joblib | 1.4.2 | Model serialization | | nltk | 3.8.1 | NLP utilities | --- ## 🌐 API Reference ### `GET /api/health` Returns system health and model status. ### `GET /api/model-info` Returns classifier metadata and feature importances. ### `POST /api/analyze` Main analysis endpoint. **Request:** ```json { "age": 68, "symptoms_score": 9, "vaccinated": false, "opinion_text": "I refuse vaccines." } ``` **Response:** ```json { "ml_prediction": { "covid_status": "Positive", "confidence": 74.04, "model_accuracy": 84.58 }, "sentiment_analysis": { "label": "Negative", "compound_score": -0.5423 }, "knowledge_inference": { "risk_level": "High Risk", "rule_fired": "R1", "reasoning_chain": "RULE R1 FIRED: ..." }, "treatment_recommendation": { "primary_recommendation": "Immediate Medical Consultation", "urgency_level": "CRITICAL" } } ``` # salaar