Song Recommendation Engine

Overview

Song Recommendation Engine is an innovative project that leverages neural networks to create personalized song recommendations. The system is designed to learn user preferences from a custom dataset by mapping song embeddings to user feedback. Here's how it works:

1. Training on Custom Dataset:
   The neural network is pre-trained on a custom dataset to understand song embeddings and their relationships to user preferences.

2. Few-Shot Learning for Personalization:
   For each new user, the program performs a few-shot learning process:

   • The user provides feedback on 5 songs they like and 5 songs they dislike.
   • The pre-trained model is fine-tuned using these preferences to better understand the user’s taste.

3. Personalized Recommendations:

   • The model generates a sigmoid score to predict the likelihood of the user enjoying a given song.
   • Songs with embeddings similar to the user’s 5 liked songs are presented to the model.
   • If the model predicts the user will like a song, it outputs a score of 1.

This dynamic process ensures that the recommendation engine adapts to individual preferences, offering tailored song suggestions based on user feedback.

Project Structure

├── src/           # Source code in .ipynb format
├── data/          # Storage of all the custom datasets and temporary json files
├── embeddings     # Directory of all the saved vector embeddings
├── utils          # Folder of all auxiliary sources
├── README.md      # Project documentation

Running Instructions

1. Install Dependencies:
   Ensure all prerequisites are installed as outlined in the prerequisites section.

   python -m pip install --upgrade pip
   pip install pytorch
   pip install numpy
   pip install sklearn
   

2. Run a program: Use the following command to test a workflow using Promptflow:

   python main2.py

If you've converted the notebook to a Python script, use the command above. Otherwise, open the Jupyter Notebook and run all the cells manually.