How Personalized Music Recommendations Can Improve Social Media Videos

After uploading video clips to TikTok or other social media, people usually keep an eye on how many likes, views, and comments they’re getting. The more, the better.

And social media platforms are regularly reviewing and adapting their algorithms to help boost engagement and interaction among users. Again: the more, the better.

With this in mind, a University of Minnesota researcher helped create a new model of background music recommendation for short videos. Unlike existing models that may only consider a user’s music preferences, this new model aims to provide highly personalized music recommendations that consider both user preferences and video themes in relation to music. Ultimately, better music recommendations could help a video get more attention and engagement.

Xuan Bi, assistant professor of Information and Decision Sciences at the Carlson School of Management, worked with four co-authors to publish a paper detailing their music recommendation model in Information Systems Research.

“We are the first to create a model that can use all the user preference, music and lyrics, and video information all together in order to make a music recommendation for video clips,” Bi said. “Conventionally, platforms only use user-music alignments rather than the alignments with music and video. We make a more comprehensive type of recommendation.”

Bi and co-authors conducted experiments with data from Douyin, a popular short-video sharing platform in China, to test their background music recommendation model. They analyzed the performance of 4,000 music clips used in 6.7 million Douyin videos created by 4.9 million users, applying their new model and comparing it to existing models.

The researchers found that their model significantly outperformed existing models by a few measures, including average likes.

These findings represent a win-win: Platforms can update or improve their algorithms while users can benefit from faster, better music choices, Bi said.

“Nowadays, platforms like Douyin and TikTok, they have their own personalized music recommendation systems already—but hopefully the work that we did can provide additional insight for them to improve their models in the future and commercialize it to create more value for the user, improving user engagement,” Bi said. “You can imagine, if the music recommendations are getting more and more accurate and the music recommended to you is precisely what you like for your video clips, user engagement can be improved to the greatest extent.”