A novel way to classify music for the purposes of archiving, sorting and music recommendation has been developed by Yan Yang of the Department of Music and Dance at Hunan University of Science and Engineering in Yongzhou, China. They publish details in the International Journal of Networking and Virtual Organisations. Their approach can assist with personalised music recommendation by employing a hybrid model based on a user attention mechanism and multi-layer memory to discern the type of musical emotion present in the music and listener behaviour data.
Very few musical artists receive the epithet – genre-defying. Music of all kinds commonly fits into categories albeit some narrower than others whether one is discussing the various forms of classical music, pop and rock, dance music, so-called world music or any other classification. Audiences often place a song into one of a handful of genres they know, while die-hard enthusiasts of a particular genre can discern numerous sub-genres within each category. In today’s era of digital music distribution, archiving, and recommendation systems, it would be useful to automate the task of classifying music into genres.
The multi-layer component of the model, which looks at long- and short-term user music preferences can glean what listeners have liked historically but also their current preferences. It combines this information with an attention mechanism to analyze the emotional attributes of the music with which users interact the most.
Yang demonstrated a recall accuracy of almost 98 percent on two different test datasets. These results indicate that the model can provide highly accurate and tailored music recommendations for users. Such an approach will hopefully benefit listeners who will have an improved experience of using a music streaming service, but it will also benefit the business of the streaming service and potentially the artists providing the musical content for those services.
Many different factors affect a listener’s music choices at a given time or in a given situation. However, when examining personalized recommendations from the vantage point of behavioural traits and emotional attributes it is possible to find ways to improve the listening experience by homing in on particular pieces of music that will be well received by the listener given their immediate environment, social connections, and other factors.
Yang, Y. (2023) ‘Research on long- and short-term music preference recommendation method integrating music emotional attention’, Int. J. Networking and Virtual Organisations, Vol. 28, Nos. 2/3/4, pp.381–397.