🎵 Music Recommendation System

A graph-based music recommendation engine that uses Node2Vec embeddings and K-Nearest Neighbors to suggest similar songs across multiple query types — by song title, artist, genre, decade, country, or year.

---

⚙️ How It Works

1. Data Preprocessing — Loads and merges music/artist datasets, removes missing values, normalizes numeric features, and enriches each song with a geographic country (derived from artist lat/lon coordinates via GeoJSON lookup).
2. Knowledge Graph Construction — Builds an undirected NetworkX graph where 1,000 song nodes connect to 7 decade hub nodes (1950s–2010s). Numeric attributes (tempo, loudness, duration, etc.) are stored as node features after min-max normalization.
3. Node2Vec Embedding — Runs biased random walks on the graph to produce 64-dimensional vector representations for every node (songs + decade hubs).
4. KNN Model — Trains a K-Nearest Neighbors model (k=6) on the embeddings to identify the most similar songs in embedding space.
5. Interactive Query Interface — Accepts a query type and value, finds matching songs in the dataset, and returns the 6 most similar recommendations.

---

🔍 Query Types

#	Query Type	Example value
1	Song title	"Hotel California"
2	Artist name	"The Beatles"
3	Decade	"1990s"
4	Country	"United States"
5	Term (genre)	"rock"
6	Year	"1975"

---

📁 Project Structure

Music-Recommendation-System/
├── Code/
│   └── Music_Recommendation_System.ipynb   # Full pipeline: preprocessing → graph → embeddings → recommendations
├── Music Data/
│   └── music_dataset.csv                   # Processed dataset (1,000 songs, 13 columns)
├── models/
│   ├── embedding                           # Pre-trained 64-dim Node2Vec embeddings (Word2Vec format)
│   └── node2vec_model                      # Saved Node2Vec model
└── .gitignore

---

📊 Dataset

The processed dataset (music_dataset.csv) contains 1,000 songs with the following columns:

Column	Description
song_title	Song name
year	Release year
release	Album / release name
tempo	Beats per minute
loudness	Audio loudness (dB)
duration	Length in seconds
song_hotttnesss	Song popularity score (0–1)
artist_name	Artist or band name
artist_hotttnesss	Artist popularity score (0–1)
artist_familiarity	How widely known the artist is (0–1)
term	Musical genre / style
country	Artist's country of origin
decade	Decade of release (e.g. 1990s)

---

🕸️ Knowledge Graph

• 1,007 nodes — 1,000 song nodes + 7 decade nodes
• 1,000 edges — each song connected to its decade hub
• Numeric features are min-max normalized before embedding to prevent any single feature from dominating distances

---

🧠 Node2Vec Configuration

Parameter	Value
Dimensions	64
Walk length	10
Walks per node	100
Window size	10
Workers	4

---

📦 Installation

Install with:

pip install shapely node2vec networkx numpy pandas gensim scikit-learn scipy matplotlib tqdm smart-open

---

🚀 Getting Started

1. Clone the repository:

   git clone https://github.com/mega5800/Music-Recommendation-System.git
   cd Music-Recommendation-System

2. Install dependencies (see above).

3. Open and run the notebook:

   jupyter notebook Code/Music_Recommendation_System.ipynb

4. The notebook will:

   • Load the preprocessed dataset from Music Data/music_dataset.csv
   • Reconstruct the knowledge graph
   • Load (or re-train) the Node2Vec embeddings from models/
   • Launch an interactive prompt where you can enter a query type and value to receive song recommendations

---