🎵 Multilingual Content-Based Music Recommendation System

A project for PRML at IIT Jodhpur that delivers personalized song recommendations based on audio features using unsupervised learning techniques like Kernel PCA and KMeans.

🧠 Project Overview

This system recommends songs in five different languages—Hindi, Tamil, Korean, English, and a miscellaneous category—based on the user's input of a song and preferred language.

By analyzing the input song's audio features, we find the most similar songs within the same language group using:

• Dimensionality Reduction (Kernel PCA)
• Clustering (K-Means with Elbow Method and KMeans++ initialization)
• Cosine Similarity for Recommendations

---

📁 Dataset

• Source: Spotify Tracks Dataset on Kaggle
• Size: 61,711 tracks
• Languages: Hindi, Tamil, Korean, English, Misc
• Key Features: tempo, valence, energy, acousticness, danceability, year, popularity, loudness, etc.

---

⚙️ Data Preprocessing

• Removed non-audio columns (e.g., artwork, album name)
• Dropped missing/duplicate rows
• One-hot encoded categorical features (key, time_signature)
• Standardized numerical features
• Grouped and saved songs by language

---

🔍 Dimensionality Reduction

• Applied Kernel PCA (RBF kernel) for capturing non-linear patterns
• Compared with PCA — Kernel PCA yielded better visual separability

---

🧩 Clustering

✅ K-Means Clustering

• Custom implementation from scratch
• KMeans++ for better centroid initialization
• Elbow Method to find optimal k per language
• Dunn Index used for validation
• Final models saved for each language group

❌ DBSCAN

• Tried with multiple epsilon values
• Poor cluster formation, most points marked as outliers
• K-Means chosen over DBSCAN for better performance

---

🎧 Recommendation Engine

• For a given input song, identify its cluster
• Use cosine similarity to recommend top-n most similar songs within the same cluster
• Metadata (track name, artist, URL, artwork) retrieved for each recommendation

---

🧪 Results

• Provided consistent and meaningful recommendations across different language groups
• Evaluated qualitatively with positive alignment to user preferences in mood, rhythm, and genre

---

🚀 How to Run the Project

1. Clone the Repository

   git clone <repo-url>
   cd MusicApp
   

2. Start the Frontend

   cd frontend
   npm install
   npm run dev
   

3. Start the Backend (in a second terminal)

   python app.py
   

   Make sure to install necessary Python imports using:

   pip install -r requirements.txt
   

4. Use the App

   • Open your browser and go to http://localhost:3000(localhost where the frontend runs)
   • Click the "Get Started" button
   • Select a language and search for a song
   • Click "Show Recommendations" to view similar songs

---

🚀 Future Work

• Use lyrics-based features
• Integrate deep learning models for better embeddings
• Add user feedback loop for more personalization

---

👥 Contributors

• Vaibhav Garg
• Saher Dev
• Swayam
• Arnav Kataria
• Tanisha Sonkar

---

🔗 References

• Scikit-learn
• Spotify Web API
• KMeans From Scratch - Medium
• IBM: K-Means Explained

---

📌 Repository

📂 GitHub Repo