The Feature Importance Analysis of Music Recommendation System with K-means and K-Nearest Neighbors

Abstract

With the rapid development of science and technology, society has entered an era of high informatization. The recommendation system can alleviate the problem of information overload due to the vast and complex information on the Internet. Music recommendation is one of the main application fields of the recommendation system. This paper revolves around building a music recommendation system using Spotify’s dataset. There are two main methods used in this paper to analyze the importance of features in music recommendation based on machine learning techniques. Specifically, this paper uses K-means clustering to identify similarities in the feature combination, bringing together songs with similar types. K-Nearest Neighbors (KNN) is used to find the nearest neighbours to a song by combining the selected features. In the evaluation part, to ensure the feature pairs are significant to KNN models, the accuracy is calculated and compared before and after one specific feature combination is removed. The results reveal that energy and valence are the most compelling feature combinations according to the cluster analysis. Besides, the accuracy after removing any feature is smaller than the accuracy using all features, which reflects that these features are essential for the KNN model. However, the feature combination (energy and valence) shows that the contradictory results indicate that any single analysis method for determining the importance of a feature is one-sided.