As digital music platforms continue to expand their catalogs, listeners face an increasing challenge: finding songs that fit their current mood. Millions of tracks are readily available through streaming services such as Spotify, but selecting the right song often depends on more than popularity or listening history. Many users search for music that helps them relax, concentrate, stay motivated, or simply reflect their emotions.
The research from Universitas Teknologi Yogyakarta addresses this challenge by introducing a recommendation system that interprets the emotional characteristics of songs based on measurable audio features. Rather than attempting to predict a listener's psychological state, the system estimates the dominant mood expressed by each song and recommends music that closely matches the mood selected by the user.
Why Mood-Based Music Recommendation Matters
Conventional music recommendation systems typically rely on user behavior, listening history, or collaborative filtering. While effective, these approaches often struggle when users have limited listening data, a situation commonly known as the "cold-start" problem.
The UTY researchers adopted a content-based recommendation approach, allowing the system to analyze songs directly through their audio characteristics instead of depending on previous user interactions. This makes the recommendation engine useful even for first-time users.
The growing demand for personalized digital experiences has made emotion-aware recommendation systems increasingly important across entertainment platforms. Transparent AI models are also gaining attention because users and developers alike want to understand how recommendations are generated rather than accepting decisions from opaque "black-box" algorithms.
A Simple and Explainable Research Approach
The researchers designed an applied quantitative study using the publicly available SpotifyFeatures.csv dataset containing 232,725 songs. Instead of processing every available audio characteristic, the model focuses on four essential Spotify audio features:
- Valence, representing the positivity or negativity of a song.
- Energy, indicating musical intensity.
- Tempo, measuring the speed of the music.
- Mode, distinguishing between major and minor tonal structures.
These four variables are processed using zero-order Sugeno Fuzzy Logic, a rule-based artificial intelligence method that is both computationally efficient and easy to interpret. The model classifies every song into one of five mood categories:
- Very Sad
- Sad
- Neutral
- Happy
- Very Happy
Each category receives a numerical score from one to five. Songs are then ranked according to how closely their scores match the listener's selected mood. Unlike many deep learning systems, every recommendation can be traced back to explicit fuzzy rules, making the decision-making process transparent.
Key Findings
The study demonstrates that the prototype successfully performs several core functions:
- It analyzes Spotify audio features to estimate the dominant mood of individual songs.
- It classifies songs into five understandable mood categories.
- It ranks songs according to their similarity to the user's desired mood.
- It provides song search functionality that predicts the mood of existing tracks.
- It delivers recommendations through a web application built with a FastAPI backend and a React frontend.
The recommendation system presents users with song titles, artist names, genres, Sugeno mood scores, predicted mood labels, and direct YouTube search links, creating an accessible interface for music exploration.
Transparent AI for Better User Trust
One of the most significant contributions of the research is its emphasis on explainability. While modern deep learning models often achieve high predictive accuracy, their internal decision processes are difficult to understand.
The Sugeno Fuzzy Logic model offers an alternative by allowing every recommendation to be explained through visible membership functions and predefined fuzzy rules.
As explained by Anandava Eka Buana Baskara and colleagues from Universitas Teknologi Yogyakarta, the prototype prioritizes interpretability over algorithmic complexity, making it particularly suitable for educational applications, practical computing projects, and transparent recommendation systems.
The researchers also carefully distinguish between song mood and listener mood. The system estimates the emotional tendency of music based solely on audio characteristics and does not attempt to detect or infer the psychological condition of users.
Practical Impact and Future Opportunities
The findings have implications beyond music streaming platforms. Transparent recommendation systems like this could support educational technology, multimedia applications, digital entertainment services, and research on explainable artificial intelligence.
Because the model does not require extensive user histories, it can also serve as a practical solution for newly launched music platforms or applications with limited behavioral data.
The researchers acknowledge several limitations. The fuzzy rules were manually designed rather than optimized using human mood evaluations, and only four Spotify audio features were included in the current model. Future versions could incorporate additional variables such as danceability, acousticness, loudness, genre, and lyrics sentiment to improve recommendation quality.
The team also recommends validating the predicted mood categories through listener studies and comparing the Sugeno model with modern machine learning approaches. Hybrid recommendation systems that combine fuzzy logic with user preference histories may further enhance personalization while maintaining explainability.
Author Profile
Anandava Eka Buana Baskara is the corresponding author of the study, conducted together with Faiz Ahmad Fauzan, Rizka Octa Setiani, Cintiya, Imiel Ardhanenggar Tallane, and Fadil Indra Sanjaya. The research team is affiliated with the Informatics Program, Universitas Teknologi Yogyakarta (UTY), Indonesia. Their academic expertise includes artificial intelligence, fuzzy logic, recommendation systems, web application development, and data mining. The published article lists Anandava Eka Buana Baskara as the corresponding author; individual academic degrees are not specified in the journal.
Source
Article Title: Mood-Based Song Recommendation System Using Sugeno Fuzzy Logic
Authors: Anandava Eka Buana Baskara, Faiz Ahmad Fauzan, Rizka Octa Setiani, Cintiya, Imiel Ardhanenggar Tallane, and Fadil Indra Sanjaya
Journal: Formosa Journal of Multidisciplinary Research (FJMR)
Volume 5, Issue 6 (2026)
0 Komentar