An Order-Complexity Aesthetic Assessment Model for Aesthetic-aware Music Recommendation

TL;DR

This work tackles the challenge of evaluating music aesthetics objectively and integrating such evaluations into music recommendations. It adopts Birkhoff's order–complexity measure and builds a four-feature aesthetic model (harmony, symmetry, chaos, redundancy) to score music, then incorporates these features into a Transformer-based recommendation backbone (CL4SRec) for sequential suggestions. Empirical results show strong discrimination in aesthetic scores (accuracy ~92%) and consistent gains in next-item prediction when aesthetic signals are included, supported by subjective listening tests. The approach offers a practical path to align music recommendations with artistic value, potentially enhancing user satisfaction in AI-generated and curated music systems.

Abstract

Computational aesthetic evaluation has made remarkable contribution to visual art works, but its application to music is still rare. Currently, subjective evaluation is still the most effective form of evaluating artistic works. However, subjective evaluation of artistic works will consume a lot of human and material resources. The popular AI generated content (AIGC) tasks nowadays have flooded all industries, and music is no exception. While compared to music produced by humans, AI generated music still sounds mechanical, monotonous, and lacks aesthetic appeal. Due to the lack of music datasets with rating annotations, we have to choose traditional aesthetic equations to objectively measure the beauty of music. In order to improve the quality of AI music generation and further guide computer music production, synthesis, recommendation and other tasks, we use Birkhoff's aesthetic measure to design a aesthetic model, objectively measuring the aesthetic beauty of music, and form a recommendation list according to the aesthetic feeling of music. Experiments show that our objective aesthetic model and recommendation method are effective.

Figures (6)

• Figure 1: We construct a music recommendation list based on the results of the proposed music aesthetic model by adding four aesthetic features into music recommendation task.
• Figure 2: Three steps from creation to hearing for a music art work, and this paper focuses on the sound stage.
• Figure 3: The overview of our aesthetic approach. We align and preprocess the midi, audio, and mel spectrogram of music, and obtain the values of ten basic music features by calculating their basic attributes. Then we put the corresponding basic music features to four logistic regression models to extract four music aesthetic features. Finally, we take the values of four music aesthetic features as the input of Birkhoff O/C model, and solve the model to get the final aesthetic score.
• Figure 4: The overview of our recommendation approach. In addition to traditional positional features and music features, we will also extract good aesthetic features as important reference standards for sequential music recommendations. Moreover, we also consider the aesthetic rating score of music as a label value in the recommendation of sequential music.
• Figure 5: The first two rows in the figure show the distribution of ten basic music features on three sample sets. The last graph in the third row shows the aesthetic score distribution of our trained model on the three sample sets. The remaining four graphs in the third row show the distribution of four aesthetic features on the three sample sets.