DataRobot Predicts the Grammy Awards’ Song of the Year Is…

Xconomy Boston — 

With everyone worrying that a robot or artificial intelligence model will take their jobs someday, why shouldn’t The Recording Academy fear its annual task of doling out the Grammy Awards is safe?

DataRobot, a Boston-based startup whose machine-learning platform helps companies use algorithms for data analysis without hiring a data science team, is taking a first swipe at predicting how likely it is that each nominee for Song of the Year will walk away with that gramophone award.

The startup, founded in 2012 by a group of data scientists at Travelers Insurance, said in an Xconomy interview last year it has already swatted away a number of acquisition offers and is playing offense to keep ahead of competitors. It raised $100 million in October, pushing its total venture funding to $225 million.

DataRobot combined data from Spotify’s API (application programming interface) for the current nominees as well as all the nominees and winning songs since the first awards ceremony in 1959. DataRobot’s Taylor Larkin wrote in a blog post that the data included “genre of the song, amount of profanity, general sentiment, total word count in the song, and various audio features derived by Spotify.”

These audio factors are listed by the music service as duration, key, time signature, “acousticness,” “danceability,” “energy,” “instrumentalness,” “liveliness,” “loudness,” “speechiness,” “valence,” and “tempo.” (Of course, anyone can make predictions—we’ll see how accurate it turns out to be.)

Without further ado, here is the (likely) winner, in decreasing order of probability:

• This is America, by Childish Gambino (with a 20.38 percent chance of winning).

• Shallow, by Lady Gaga and Bradley Cooper (19.17 percent)

• All the Stars, by Kendrick Lamar, SZA (17.21 percent)

• The Middle, by Zedd, Maren Morris, Grey (16.47 percent)

• The Joke, by Brandi Carlile (15.55 percent)

• God’s Plan, by Drake (15.48 percent)

• Boo’d Up, by Ella Mai (14.54 percent)

• In My Blood, by Shawn Mendes (with 13.08 percent)

(One wonders how the fact that Childish Gambino, Drake, and Kendrick Lamar turned down offers to perform at the Grammy’s would impact their likelihood of winning, but that might be beyond the abilities of a machine learning model.)

Larkin, a data science evangelist at DataRobot, wrote in October about the Spotify API; he built a model to find which song factors are related to song popularity. (You can read more of that here.) Interestingly, a modest amount of profanity was one driving factor.