Simplest approach is popularity-based recommender in which we first train a popularity_model and then we recommend song to the users based on how much that song is popular or simply how many times that song is being listened. But its limitation is that it has no personalization.
Another approach can be, we can train a personalized_model then we can recommend the user similar artist's songs may be.
Then I compared the model performance of both popularity_model and personalized_model