Item-based Collaborative Filtering for Music Recommender System

This post details the implementation of an item-based collaborative filtering (I2I CF) recommender system for music, leveraging Milvus as the embedding similarity search engine. It describes the process of extracting song feature embeddings using CNNs from Mel-frequency cepstrum (MFC) and deploying Milvus in a highly available cluster configuration with Mishards on Kubernetes for production. Additionally, Milvus is used as a preliminary step for duplicate song filtering by searching for similar audio fingerprints.