Mood-Based Song Recommendation System Using TF-IDF and Cosine Similarity

Andri Riyan Dwi Putra Siregar (Muhammadiyah University of North Sumatra Medan)
Rizaldy Khair (Muhammadiyah University of North Sumatra Medan)

Journal of Scientific Research, Education, and Technology
Vol. 5 No. 4 (2026): Vol. 5 No. 4 2026

This study created a mood-based music recommendation system using a content-based filtering approach with TF-IDF and Cosine Similarity to handle the huge amount of digital music data. Using about 15,000 songs from Kaggle, the system processes metadata and audio features like valence, energy, danceability, and tempo, all implemented in a Python and Flask web app. The mix of text similarity and audio features works really well for giving spot-on recommendations, as shown by performance scores of 0.97 accuracy, 0.90 precision, 0.82 recall, and 0.86 F1-score. Plus, with a System Usability Scale (SUS) rating in the good range, the system is not just technically accurate but also super easy to use and well-liked by users.