Music Recommendation System Based on Spotify Audio Features: Collaborative Application of KNN, K-Means, and DBSCAN

This article introduces a music recommendation system built using Spotify audio features and three unsupervised learning algorithms—KNN, K-Means, and DBSCAN—exploring the technical implementation of clustering and similarity calculation in personalized recommendation. The system combines the advantages of different algorithms to provide recommendation capabilities for cold-start users and new songs, while supporting style discovery and niche music mining.

In the era of digital music streaming, recommendation systems are the bridge connecting users to massive music libraries. Compared to collaborative filtering, audio content-based recommendation algorithms do not rely on user historical behavior data; they can directly analyze music features and provide recommendation capabilities for cold-start users and newly released songs.

This project was developed by amperry01. The core data source comes from the audio feature API of the Spotify platform, including quantitative indicators such as tempo, energy, danceability, loudness, key, instrumentalness, liveness, time signature, speechiness, and valence. These indicators allow music of different styles to be compared and analyzed in the same feature space.

The KNN algorithm selects the nearest K songs as recommendation results by calculating the Euclidean distance or cosine similarity between the target song and all songs in the library. Its advantages are simple implementation and strong interpretability, and it can capture local similarity. However, its computational complexity increases with the amount of data, requiring index optimization.

K-Means divides songs into clusters based on audio feature similarity, representing potential styles or emotional types. It automatically discovers the internal structure of data and can be used to generate similar style recommendations or as a dimensionality reduction preprocessing step to improve the efficiency of subsequent algorithms.

DBSCAN does not require pre-specifying the number of clusters; it can identify clusters of arbitrary shapes and detect noise points. It can discover unique sub-genres within mainstream styles, filter abnormal tracks, and has important value for niche music and long-tail content mining.

The project applies three algorithms simultaneously: KNN provides precise single-point queries, K-Means realizes global library organization, and DBSCAN is responsible for anomaly detection and subgroup discovery. Combining these builds a robust and comprehensive recommendation service to meet the needs of different scenarios.

Audio feature-based recommendation is an important direction for content understanding-based recommendation. In the future, it is expected to combine traditional features with deep representations extracted by deep learning to achieve a more accurate and personalized music discovery experience.