📊 Employee Sentiment Analysis — LLM & NLP Pipeline

An end-to-end NLP pipeline that analyzes employee email feedback to detect sentiment, rank employees, identify flight risks, and model sentiment trends over time using TextBlob and Scikit-learn.

---

✨ What This Project Does

• 🏷️ Sentiment Labelling — TextBlob polarity scoring maps each feedback to Positive / Neutral / Negative
• 📈 Monthly Sentiment Scoring — Aggregates average sentiment over time with a line chart
• 🏆 Employee Ranking — Ranks all employees by their average sentiment score
• ⚠️ Flight Risk Detection — Rule-based heuristic flags high-risk employees (avg sentiment ≤ -0.1 with ≥ 3 feedbacks)
• 📉 Linear Regression Trend — Fits a trend line over monthly sentiment to detect improving or declining patterns
• 🔍 EDA & Visualizations — Text length distributions, feedback counts per year, sentiment distribution charts

---

🛠️ Tech Stack

Layer	Technology
Language	Python 3.9+
Sentiment Engine	TextBlob
Data Processing	Pandas, NumPy
ML / Trend Model	Scikit-learn (LinearRegression)
Visualizations	Matplotlib, Seaborn
Notebook	Jupyter
Config	python-dotenv

---

📁 Project Structure

sentiment-analysis-LLM/
├── data/
│   └── employee_feedback.csv       # Email dataset (included)
├── notebooks/
│   └── employee_sentiment_analysis.ipynb  # Full interactive pipeline
├── src/
│   └── sentiment_pipeline.py       # Standalone Python script
├── outputs/                        # Generated on run
│   ├── feedback_with_sentiment.csv
│   ├── employee_ranking_and_flight_risk.csv
│   └── monthly_sentiment_trend.csv
├── .env.example
├── requirements.txt
└── README.md

---

📂 Dataset

The dataset data/employee_feedback.csv contains employee email data with these columns:

Column	Used As	Description
from	employee_id	Sender email (employee identifier)
body	feedback_text	Email content (analyzed for sentiment)
date	date	Date of the feedback
Subject	—	Email subject (kept for reference)

---

⚙️ Setup & Installation

1. Clone the repository

git clone https://github.com/nithu0035/sentiment-analysis-LLM.git
cd sentiment-analysis-LLM

2. Create a virtual environment (recommended)

python -m venv venv

# Windows:
venv\Scripts\activate

# macOS / Linux:
source venv/bin/activate

3. Install dependencies

pip install -r requirements.txt

4. Configure environment variables

cp .env.example .env

.env should contain:

DATA_PATH=./data/employee_feedback.csv
OUTPUT_DIR=./outputs

---

▶️ Usage

Option A — Jupyter Notebook (recommended)

jupyter notebook

Open notebooks/employee_sentiment_analysis.ipynb and run all cells (Kernel → Restart & Run All).

Option B — Python Script

python src/sentiment_pipeline.py

Both options produce the same 3 output CSVs saved to the outputs/ folder.

---

🧠 Methodology

1. Data Cleaning — Parse dates, drop nulls, rename columns, add text length feature.

2. Sentiment Analysis — TextBlob polarity score in range [-1, 1] mapped to three labels:

• score > 0.05 → Positive
• score < -0.05 → Negative
• otherwise → Neutral

3. Monthly Aggregation — Group by year-month, compute average sentiment and feedback count, plot time-series.

4. Employee Ranking — Group by employee ID, rank by average sentiment score descending.

5. Flight Risk Flag — Employees with avg sentiment ≤ -0.1 AND ≥ 3 feedback entries are flagged as High risk.

6. Linear Regression Trend — Encodes months as integer time index, fits LinearRegression, plots actual vs predicted trend line. A positive slope = improving sentiment over time.

---

📤 Output Files

File	Description
feedback_with_sentiment.csv	All feedback rows with polarity score and label
employee_ranking_and_flight_risk.csv	Per-employee avg sentiment, feedback count, rank, and flight risk flag
monthly_sentiment_trend.csv	Monthly avg sentiment with linear regression predictions

---

📄 License

MIT

👤 Author

Gudipatoju Nitesh
GitHub: @nithu0035