🎵 digital-music-hybrid-recommender

This project implements a hybrid recommender system trained on the Amazon Reviews 2023 Digital Music dataset by McAuley Lab. It integrates content-based (TF-IDF + cosine), collaborative (item-based + SVD), and popularity-based (Bayesian weighted) recommendation techniques to produce personalized digital music recommendations.

The system provides an interactive Streamlit web interface that allows users to upload .jsonl review and metadata files, explore top-rated songs or albums, and adjust model weights dynamically to balance content, collaborative, and popularity signals.

Built for reproducible RecSys benchmarking, it features:

• Fine-grained data cleaning and merging of reviews and metadata

• TF-IDF text vectorization on titles, categories, and descriptions

• Sparse cosine similarity and latent SVD embeddings for collaborative filtering

• Weighted hybrid blending with user-adjustable sliders

• Tabbed visualization for Popular, Content-based, Collaborative, and Hybrid recommendations

  Excellent — I’ve reviewed your uploaded Time_Series_Model.py script. It’s a Streamlit-based hybrid recommender system (Content-based + Collaborative + Popularity + Hybrid blend) built for the Amazon Digital Music dataset (2023) from McAuley Lab.

---

🎵 Digital Music Hybrid Recommender System

A hybrid recommendation system for Amazon Digital Music Reviews (2023), built using Streamlit, TF-IDF, and Collaborative Filtering.
This project blends popularity, content-based, and collaborative approaches to generate personalized digital music recommendations.

---

📘 Dataset

This model uses the Amazon Reviews 2023 — Digital Music dataset from McAuley Lab:

Feature	Description
Reviews	571M+ user-item interactions (May 1996 – Sep 2023)
Metadata	Rich item info (title, category, description, etc.)
Granularity	Timestamps at the second-level
Splits	Standard RecSys train/test splits

---

⚙️ Features

✅ Load and clean .jsonl reviews and metadata
✅ TF–IDF based content similarity model
✅ Item–item collaborative filtering (cosine + SVD)
✅ Popularity (Bayesian weighted mean) ranking
✅ Hybrid weighted recommender (configurable weights)
✅ Fully interactive Streamlit interface
✅ User-controlled tuning sliders and data uploads

---

🧠 Model Architecture

Module	Technique	Description
Popularity	Bayesian average	Weighted rating using vote counts
Content-based	TF-IDF + Cosine similarity	Textual similarity on title + category + description
Collaborative	Sparse cosine NN + SVD	Item–item matrix factorization
Hybrid	Weighted fusion	Combines all scores via linear weights

---

🧩 Tech Stack

• Python 3.9+
• Streamlit for deployment/UI
• scikit-learn (TF-IDF, SVD, NearestNeighbors)
• pandas, NumPy, SciPy
• McAuley Lab Amazon Dataset (2023)

---

Launch the app

streamlit run Time_Series_Model.py

Upload Data

You can:

• Use default file paths (if placed in data/)
• Or upload .jsonl review and meta files via the Streamlit sidebar

---

🧪 Example Weights for Hybrid Blend

Component	Default Weight
Content-based	0.45
Collaborative	0.45
Popularity	0.10

---

📈 Output Tabs

Tab	Description
Popular	Top-N Bayesian weighted items
Content-based	TF-IDF cosine recommendations
Collaborative	Item–item cosine + SVD latent factors
Hybrid	Final blended recommendations

---

🧾 Citation

If you use this dataset or pipeline, please cite:

He, Ruining, McAuley, Julian. Amazon Product Data 2023 (McAuley Lab, UCSD). 📑 Paper

---

📸 Demo Screenshot

---

🧰 Folder Structure

.
├── Time_Series_Model.py     # Streamlit app
├── requirements.txt
├── data/                    # Example .jsonl data
├── assets/                  # Screenshots or demo
├── notebooks/               # Optional exploration
└── models/                  # Saved models (TF-IDF, SVD, etc.)

---

🪄 Future Improvements

• ✅ Integrate HuggingFace embeddings for better semantic similarity
• ✅ Add session-based or sequence models (e.g., SASRec, BERT4Rec)
• 🔜 Deploy on Streamlit Cloud or HuggingFace Spaces
• 🔜 Add user profiling and A/B testing

---

📜 License

This project is licensed under the MIT License.

---

👨‍💻 Author

Venkatesh Data Science | Machine Learning | Recommender Systems 📧 venkateshvarada56@gmail.com 🌐 LinkedIn Profile