NLP-Sentiment-Analysis

🧠 Sentiment Analysis Platform

A Python NLP system that classifies text as positive, negative, or neutral, detects 8 emotions, and auto-generates a visual HTML report — powered by a VADER + TextBlob ensemble.

---

📊 Real Results (from 36 analysed texts)

Overall Sentiment

Sentiment	Count	Percentage
✅ Positive	20	55.6%
❌ Negative	13	36.1%
➖ Neutral	3	8.3%
Total	36	100%

Sentiment by Source

Source	Positive	Negative	Neutral
Amazon Reviews	7	5	0
Social Media	7	4	1
News Articles	6	4	2

Model Confidence

Metric	Score
Average confidence	79.4%
Highest confidence	100.0%
Lowest confidence	56.2%

Emotion Detection

Emotion	Avg Intensity	Count as Dominant
😄 Joy	0.1458	6
😲 Surprise	0.1458	5
😠 Anger	0.1389	6
😨 Fear	0.0833	3
🤝 Trust	0.0694	2
😢 Sadness	0.0417	1
🔮 Anticipation	0.0278	1
🤢 Disgust	0.0139	0

NLP Model Scores

Model	Mean Score
VADER compound	0.1297
TextBlob polarity	0.0563
TextBlob subjectivity	0.5250

---

✨ Features

• 🔍 Sentiment classification — positive / negative / neutral using ensemble NLP
• 🎭 8-emotion detection — joy, anger, fear, sadness, surprise, trust, disgust, anticipation
• 📦 3 data sources — Amazon reviews, social media (Twitter/Instagram/Reddit), news articles
• 📈 8 auto-generated charts — radar, heatmap, word cloud, scatter plots, histograms
• 📄 Self-contained HTML report — open in any browser, no server needed
• 📁 CSV export — all scores per record, ready for Excel or Google Sheets
• 💡 Business insights — auto-derived actionable recommendations from patterns

---

🗂️ Project Structure

sentiment_analysis/
├── main.py                       # Entry point — runs full pipeline
├── data/
│   └── sample_data.py            # 36 labelled records (Amazon + social + news)
├── models/
│   └── sentiment_engine.py       # VADER + TextBlob ensemble + emotion lexicon
└── utils/
    ├── visualizer.py             # 8 matplotlib / seaborn / wordcloud charts
    └── report_generator.py       # Self-contained HTML report builder

---

⚙️ Installation

pip install vaderSentiment textblob pandas matplotlib seaborn wordcloud scikit-learn

---

🚀 Usage

python main.py

Terminal output:

────────────────────────────────────────────────
  Step 1 / 4 — Loading and analysing data
  ✓ 36 records processed
  positive    20 | negative    13 | neutral    3

  Step 2 / 4 — Generating charts
  ✓ 8 charts saved

  Step 3 / 4 — Deriving business insights
  1. Overall sentiment is strongly positive (56%)...

  Step 4 / 4 — Building HTML report
  ✓ Report saved → reports/report.html

  Done in 2.0s
────────────────────────────────────────────────

---

📁 Output Files

File	Description
reports/report.html	Full visual report — open in any browser
reports/results.csv	All scores per record
reports/charts/	8 individual PNG charts

---

➕ Add Your Own Data

Open data/sample_data.py and add entries:

{"id": "A013", "source": "amazon", "product": "Laptop",
 "text": "Incredible performance, boots in seconds. Best laptop I've owned!", "rating": 5},

Then run python main.py — the report regenerates automatically.

---

🛠️ How It Works

Ensemble Scoring

Final Score = (VADER × 0.6) + (TextBlob × 0.4)

Score ≥  0.05  →  Positive
Score ≤ -0.05  →  Negative
Otherwise      →  Neutral

• VADER — handles emoji, slang, ALL CAPS, social media tone
• TextBlob — stronger on formal and news-style writing

Emotion Detection

A custom lexicon maps words to 8 emotions. Each text receives a 0–1 score per emotion; the highest becomes the dominant emotion.

---

📦 Tech Stack

Library	Purpose
vaderSentiment	Social media sentiment scoring
TextBlob	Polarity & subjectivity
Pandas	Data processing & CSV export
Matplotlib + Seaborn	Charts & heatmaps
WordCloud	Keyword frequency visualization

---

📌 Key Insights from Sample Data

1. 56% positive overall — strong signal for marketing campaigns
2. Amazon shows 42% negative — product quality improvements recommended
3. Joy and Anger equally dominant — polarised audience, needs segmentation
4. 79.4% average model confidence — texts below 60% flagged for manual review
5. 52.5% average subjectivity — balanced mix of opinion and factual content

---

📄 License

MIT License — free to use, modify, and distribute.

---

👤 Author

shhhoaib · github.com/shhhoaib