K-Pop Girl Group Lyrics Analysis 🎤

A neural network-based classification model that predicts K-Pop girl group generations based on song lyrics analysis.

---

📋 Project Overview

This project applies machine learning techniques to analyze K-Pop girl group lyrics and predict which generation a song belongs to. The model achieves ~90% accuracy in classifying songs into five distinct generations.

Generation Classification (5-Generation Standard)

Following the reference table (韩国女团世代表_一代至五代.xlsx):

Generation	Years	Representative Artists
Gen 1 (一代)	1996-2002	S.E.S, Fin.K.L, Baby V.O.X, Jewelry
Gen 2 (二代)	2003-2009	Girls' Generation, Wonder Girls, KARA, 2NE1, f(x)
Gen 3 (三代)	2010-2013	SISTAR, Apink, EXID, Miss A, AOA
Gen 4 (四代)	2014-2017	TWICE, BLACKPINK, Red Velvet, MAMAMOO, GFriend
Gen 5 (五代)	2018+	IZ*ONE, ITZY, aespa, IVE, (G)I-DLE, NewJeans, LE SSERAFIM

---

📊 Dataset

Item	Description
Source	Kpop-lyric-datasets
Original Size	25,696 K-Pop songs from Melon Monthly Chart (2000-2023)
Filtered Dataset	3,243 girl group songs
Cleaned Dataset	~2,967 songs after data cleaning
File	girlgroup_songs.csv

Dataset Columns

• generation: Girl group generation (一代 ~ 五代)
• artist: Artist name
• song_name: Song title
• lyrics: Full lyrics text
• year, month: Chart appearance date
• rank: Chart ranking (1-100)
• lyrics_length: Character count of lyrics

---

🛠️ Technical Stack

Category	Tools
Language	Python 3.8+
ML Framework	scikit-learn
Data Processing	pandas, numpy
Visualization	matplotlib, seaborn
Text Processing	TF-IDF Vectorization

---

📁 Repository Structure

├── kpop_lyrics_analysis.py        # Main analysis script
├── girlgroup_songs.csv            # Dataset (3,243 songs)
├── model_results.png              # Visualization output
├── CA6000_Report_Final.docx       # Assignment report
└── README.md

---

🚀 Quick Start

Prerequisites

pip install pandas numpy scikit-learn matplotlib seaborn

Run Analysis

python kpop_lyrics_analysis.py

Expected Output

============================================================
K-Pop Girl Group Lyrics Analysis
Neural Network-based Generation Prediction Model
============================================================

SECTION 1: Data Import and Initial Inspection
Dataset Shape: (3243, 14)
...

SECTION 6: The Accuracy of the Eventual Model
========================================
  OVERALL TEST ACCURACY: ~90%
========================================

---

🧠 Model Architecture

Multi-Layer Perceptron (MLP) Neural Network

Input Layer (3,000 TF-IDF features)
           ↓
Hidden Layer 1 (256 neurons + ReLU)
           ↓
Hidden Layer 2 (128 neurons + ReLU)
           ↓
Hidden Layer 3 (64 neurons + ReLU)
           ↓
Output Layer (5 classes - Softmax)

Training Configuration

• Optimizer: Adam
• Early Stopping: Enabled (10% validation hold-out)
• Max Iterations: 200
• Train/Test Split: 80/20 (stratified)

---

📈 Results

Metric	Value
Test Accuracy	~90%
Macro F1-Score	~0.90
Best Validation Score	~0.95

Per-Class Performance

Generation	Precision	Recall	F1-Score
Gen 1	~0.80	~0.70	~0.75
Gen 2	~0.90	~0.95	~0.92
Gen 3	~0.90	~0.85	~0.87
Gen 4	~0.95	~0.95	~0.95
Gen 5	~0.95	~0.95	~0.95

---

🔍 Key Findings

1. Gen 4 & Gen 5 achieved highest accuracy - Modern groups have distinctive lyrical patterns with more English content
2. Gen 2 showed strong performance - Largest sample size (34%) with clear characteristics
3. Gen 1 had lowest recall - Smallest sample size (6.9%) and stylistic overlap with Gen 2 ballads
4. The model successfully learned generation-specific vocabulary and linguistic patterns

---

🤝 Acknowledgments

• Dataset: EX3exp/Kpop-lyric-datasets
• Generation reference: 韩国女团世代表_一代至五代.xlsx
• AI coding assistance: Claude (Anthropic)

---

📄 License

This project is for educational purposes as part of CA6000 coursework.