AI-based Background Music Selection System for Retail Stores



In September 2020, as a result of collaborative research conducted together with USEN CORPORATION of USEN-NEXT GROUP on the “AI BGM” project , “U MUSIC” was officially released as a product. In this project, Qosmo is in charge of providing the technology of AI-based music selection and the background visual displayed on the screen. In addition to large-scale complexes such as HANEDA INNOVATION CITY, a variety of stores and facilities have begun to use U MUSIC.


The music selection system used to be based entirely on programs (channels) curated by humans that the users at each facility could choose from. Even with a large number of channels were available, they could not satisfy the needs of all users/stores. The new music selection system is aimed at automatically customizing playlists precisely tailored to individual retail spaces. Selecting from songs distributed by USEN, the AI selects background music that changes along with the seasons, the weather, the time of day, and other factors.

The first phase consists of a prototype at The Core Kitchen/Space in Toranomon Hills, which began on July 1, 2019 and is scheduled to run for one year. Please click here for more details of proof of concept. After that, AI BGM was introduced on a trial basis at the Work Styling Nihonbashi Mitsui Tower, a corporate shared office developed by Mitsui Fudosan Co., as one of the innovative system to stimulate discussions and promote comfort and relaxation.


First, USEN provided unique image word tags such as melancholic or summer. Qosmo’s AI technology was trained on approximately 500,000 songs that had been tagged with image words, and it can automatically assign tags to new songs. In addition to image words, the AI can recognize characteristics that indicate the ambiance of a song and assign tags such as morning/evening, summer/winter, and sunny/rainy, or genres like rock/jazz. These song characteristics and the image words were combined and ranked by the AI, resulting into a score that captures a wide range of features for a single song.

Interestingly, tagging by AI goes into the “in-between” realm that is difficult to quantify for humans, such as light and sentimental at the same time.
We hope that it will lead to the selection of songs in a different order regardless of generation as well as the songs that contain unexpected elements that were not selected in the previous system.

When songs are selected, in addition to the static characteristics, real-time information such as the time of day, season, and weather can be acquired to dynamically select songs according to the changing environment. Moreover, users can “Like” or “Dislike” particular songs that will automatically be taken into consideration to generate a playlist that matches their taste.

The AI selects songs for a day by moving steadily through a group of songs distributed in higher-dimensional space, taking into account the order of song selection. With this mechanism, we aimed to achieve a natural selection of music that smoothly changes the mood of the playlist.

In this age where opportunities to come across new music are gradually diminishing due to similar recommendations, we hope “U MUSIC” will provide fresh encounters with music that you never knew existed by delivering it in a series of choices that transcend genres and generations.





  • Techical Direction

    Nao Tokui (Qosmo, Inc.)

  • Machine Learning

    Max Frenzel (Qosmo, Inc.), Bogdan Teleaga (Qosmo, Inc.)

  • Server Development

    Bogdan Teleaga (Qosmo, Inc.)

  • Background Visual

    Shoya Dozono (Qosmo, Inc.)

  • Project Management

    Hiroshi Yamato (Signal Compose, Inc.), Yumi Takahashi (Qosmo, Inc.)

Get in touch with us here!