<![CDATA[The growth of music streaming services like Spotify has encouraged users to explore millions of songs, making effective recommendation systems essential. This study examines content-based music recommendation systems by comparing feature configurations and similarity metrics. The system uses 11 Spotify audio features, both with and without genre features encoded using one-hot encoding. Similarity is calculated using cosine similarity and Euclidean distance on normalized features (MinMaxScaler) to generate the top 10 recommendations. Recommendations are considered relevant if the recommended song is by the same artist as the searched song. Performance is measured using Precision@10 and Recall@10 on 100 samples. Using audio features alone yields Precision@10 of 3.80–3.90% and Recall@10 of 6.36–6.45%. The addition of genre features improved performance to Precision@10 of 9.00–9.10% and Recall@10 of 9.06–9.89%. These results show that genre features significantly improve the relevance of recommendations, while both similarity metrics perform similarly when the features have been well normalized. This study contributes by demonstrating that feature representation plays a more critical role than similarity metrics in content-based music recommendation.]]>
Copyrights © 2026
